WO2022048288A1 - 基于地质特征约束的储层参数预测方法、装置和存储介质 - Google Patents

基于地质特征约束的储层参数预测方法、装置和存储介质 Download PDF

Info

Publication number
WO2022048288A1
WO2022048288A1 PCT/CN2021/103487 CN2021103487W WO2022048288A1 WO 2022048288 A1 WO2022048288 A1 WO 2022048288A1 CN 2021103487 W CN2021103487 W CN 2021103487W WO 2022048288 A1 WO2022048288 A1 WO 2022048288A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural network
seismic
deep neural
network model
data
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.)
Ceased
Application number
PCT/CN2021/103487
Other languages
English (en)
French (fr)
Inventor
许凯
孙振涛
王世星
唐金良
曹慧兰
张如一
郑笑雪
姚铭
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.)
China Petroleum and Chemical Corp
Sinopec Geophysical Research Institute
Original Assignee
China Petroleum and Chemical Corp
Sinopec Geophysical Research Institute
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
Priority claimed from CN202010928912.0A external-priority patent/CN114152977B/zh
Priority claimed from CN202010931370.2A external-priority patent/CN114152978B/zh
Application filed by China Petroleum and Chemical Corp, Sinopec Geophysical Research Institute filed Critical China Petroleum and Chemical Corp
Priority to US18/005,011 priority Critical patent/US20230314649A1/en
Publication of WO2022048288A1 publication Critical patent/WO2022048288A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/40—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging
    • G01V1/44—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging using generators and receivers in the same well
    • G01V1/48—Processing data
    • G01V1/50—Analysing data
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G01V1/30—Analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning

Definitions

  • the present disclosure belongs to the technical field of geophysical exploration, and in particular, relates to a method, device, computer storage medium and computer equipment for predicting reservoir parameters based on geological feature constraints.
  • this technology mainly uses depth domain "wavelet” extraction, combined with conventional inversion technology, to carry out direct prediction of elastic parameters, but the theoretical model based on depth domain data has not been established, so the basic theory is insufficient; the third is represented by Jason company
  • the depth-domain reservoir parameter prediction method based on high-precision velocity volume transformation, this method converts the depth-domain data into time-domain data for conventional reservoir parameter prediction, but the deep-time transformation has a certain transformation cumulative error, and it is time-consuming and labor-intensive. It is not conducive to improving the reservoir prediction accuracy.
  • the present disclosure proposes a method, device, computer storage medium, and computer equipment for predicting reservoir parameters based on geological feature constraints.
  • the present disclosure provides a method for predicting reservoir parameters based on geological feature constraints, comprising the following steps:
  • step S100 includes the following steps:
  • the correlation between different types of seismic attributes of the target interval and the reservoir parameters is determined through intersection analysis, and the correlation degree is selected from different types of seismic attributes that exceed the preset correlation.
  • the seismic attribute of the degree threshold is regarded as the dominant seismic attribute;
  • the data of the seismic attribute is decomposed and reconstructed by singular spectrum analysis, wherein in the reconstructed sequence, the contribution degree is kept higher than the preset contribution according to the contribution degree of the seismic attribute
  • the sequence component of the degree threshold is used as the dominant component of the dominant seismic attribute.
  • the waveform classification network model is an SOM unsupervised network model designed based on the SOM unsupervised clustering algorithm, and the network model includes a seismic attribute input layer and a classification result output layer.
  • the geological features comprise sedimentary features.
  • each of the deep neural network models is an LSTM-RNN recurrent neural network model, and the network model includes a seismic attribute input layer, a reservoir parameter output layer, and a seismic attribute input layer. and the hidden layer between the reservoir parameter output layer; wherein, the hidden layer includes:
  • the dropout layer is used to alleviate the overfitting phenomenon in the network model training process
  • regression layer which is used as the output for training the network model.
  • the step S400 further includes:
  • Different deep neural network models are trained using seismic data and logging data belonging to the target interval to optimize the model parameters of each deep neural network model.
  • the spatial variation coefficient of each trained deep neural network model in the spatial variation neural network prediction model set is determined based on the similarity of the waveforms and the spatial distance.
  • step S600 different trained deep neural network models are fused into a set of spatially varying neural network prediction models according to the following formula:
  • V p represents the reservoir parameters
  • f k (x 1 , x 2 ,...x N ) represents the deep neural network model under the kth geological feature
  • wk , i, j represent the kth geological feature under the
  • the spatial variation coefficients of the deep neural network model, x 1 , x 2 , ... x N represent different types of seismic attributes.
  • w is the spatial variation coefficient
  • v 1 is the seismic trace of the deep neural network model that has been constructed
  • v 2 is the seismic trace of the deep neural network model to be constructed
  • w c represents the seismic trace of the deep neural network model that has been constructed and the depth to be constructed.
  • the interpolation coefficient of the waveform similarity of the neural network model seismic trace w d represents the interpolation coefficient of the distance between the seismic trace of the built deep neural network model and the seismic trace of the deep neural network model to be constructed
  • c 12 represents the correlation between v 1 and v 2
  • d 12 represents the distance between v 1 and v 2
  • ⁇ is the adjustment factor
  • ⁇ c and ⁇ d are the exponential factors.
  • the above-mentioned reservoir parameters of the target interval include a spatial three-dimensional elastic parameter volume of the target interval
  • the method further includes outputting a distribution map of the spatial three-dimensional elastic parameter volume of the target interval.
  • the present disclosure also provides a reservoir parameter prediction device based on geological feature constraints, characterized in that it includes:
  • the attribute screening module is used to use the seismic data and logging data of the target interval to analyze the correlation between different types of seismic attributes of the target interval and the reservoir parameters. Select the dominant seismic attribute from the attributes;
  • the waveform classification module is used to classify the seismic waveforms of the target interval according to the waveform characteristics by using the preset waveform classification network model based on the dominant seismic attributes, and obtain the waveform classification results; wherein, different types of waveforms correspond to different geological features;
  • the model building module is used to construct different deep neural network models corresponding to different geological features with seismic attributes as input, reservoir parameters as output, and waveform classification results as constraints;
  • a model training module for using the seismic data and logging data of the target section as training data and prediction data to train the different deep neural network models to optimize the model parameters of each deep neural network model;
  • the model fusion module is used to fuse different trained deep neural network models into a set of spatially varying neural network prediction models
  • the parameter prediction module is used to predict the reservoir parameters of the target interval by using a set of spatially varying neural network prediction models.
  • the present disclosure also provides a computer storage medium, characterized in that a computer program executable by a processor is stored therein, and when the computer program is executed by the processor, the above-mentioned geological feature constraint-based algorithm is implemented. Reservoir parameter prediction method.
  • the present disclosure further provides a computer device, characterized in that it includes a memory and a processor, and the processor is configured to execute a computer program stored in the memory, so as to implement the above geological feature constraint-based algorithm. Reservoir parameter prediction method.
  • the violent video classification technology incorporating internal and external knowledge provided by the present disclosure has the following advantages or beneficial effects:
  • the present disclosure adopts the LSTM-RNN neural network model to describe the nonlinear mapping relationship between seismic attributes and reservoir parameters, which not only considers the up-down correlation of seismic data and the up-down correlation of logging data, but also takes into account the seismic data.
  • the time series characteristics of the data and logging data establish a more accurate well-seismic mapping relationship than the prior art.
  • the present disclosure corresponds to a deep network model under the same geological feature (eg, sedimentary feature) by introducing waveform clustering, and by constructing a set of spatially varying neural network prediction models, with multi-type seismic attribute data as input, and geological features as Constraints, different network models are used to predict reservoir parameters under different waveform characteristics, which effectively improves the prediction accuracy.
  • geological feature eg, sedimentary feature
  • the present disclosure is a nonlinear reservoir parameter prediction technology based on geological feature constraints, which can realize direct prediction of reservoir parameters, especially in the depth domain, and improve the prediction accuracy and spatial stability of reservoir parameters.
  • the present disclosure helps to further improve Drilling success rate, reduce oilfield exploration and development costs, and improve oilfield production efficiency.
  • FIG. 1 is a schematic flowchart of a method for predicting reservoir parameters based on geological feature constraints according to Embodiment 1 of the present disclosure
  • FIG. 2 is a schematic diagram of the SOM unsupervised clustering network model in the method for predicting reservoir parameters according to Embodiment 1 of the present disclosure
  • FIG. 3 is a schematic diagram of the LSTM-RNN cyclic neural network model in the method for predicting reservoir parameters according to Embodiment 1 of the present disclosure
  • FIG. 4 is a schematic diagram of the forget gate in the LSTM-RNN cyclic neural network model according to the first embodiment of the present disclosure
  • FIG. 5 is a schematic diagram of an incoming gate in the LSTM-RNN cyclic neural network model according to Embodiment 1 of the present disclosure
  • FIG. 6 is a schematic diagram of an output gate in the LSTM-RNN cyclic neural network model according to Embodiment 1 of the present disclosure
  • FIG. 7 is a schematic diagram of the spatial variation coefficient in the construction of the spatial variation neural network prediction model set according to the first embodiment of the present disclosure
  • FIG. 8 is a schematic diagram of waveform classification input data according to Embodiment 2 of the present disclosure.
  • FIG. 9 is a schematic diagram of the waveform classification result of the second embodiment of the present disclosure along the O72 slice;
  • FIG. 10 is a schematic diagram of the input data of the deep neural network according to the second embodiment of the present disclosure.
  • FIG. 11 is a schematic diagram of constructing a deep neural network model according to Embodiment 2 of the present disclosure.
  • FIG. 12( a ) is a schematic diagram of slices along layers of a multi-model elastic parameter prediction result based on a deep neural network according to Embodiment 2 of the present disclosure
  • Fig. 12(b) is a schematic diagram of slice along the layer of the prediction result of the elastic parameter of a single model based on the deep neural network according to the second embodiment of the present disclosure
  • Fig. 12(c) is a schematic diagram of slice slices along the slice of the time domain inversion result of the second embodiment of the present disclosure
  • Fig. 12(d) is a schematic diagram of the engineering development and deployment scheme of the second embodiment of the present disclosure.
  • Fig. 13(a) is a comparison diagram of the original logging curve of well PU_IA and the multi-model prediction result according to the second embodiment of the present disclosure
  • Fig. 13(b) is a comparison diagram of the original logging curve of well PU_IB and the multi-model prediction result according to the second embodiment of the present disclosure
  • Fig. 13(c) is a comparison diagram of the original logging curve of well PU_IC and the multi-model prediction result according to the second embodiment of the present disclosure
  • Fig. 13(d) is a comparison diagram of the original logging curve of the PU_IC well and the prediction result of the single model according to the second embodiment of the present disclosure.
  • the present disclosure proposes a depth-domain reservoir parameter direct prediction technology based on geological feature constraints, so as to improve the In particular, the prediction accuracy of reservoir parameters in the depth domain improves the stability of spatial prediction, and provides a reasonable understanding and high-precision data for subsequent drilling and reservoir simulation to support efficient exploration and development.
  • the core idea of the present disclosure is: for example, in different depositional environments, using the same model for prediction may lead to low prediction accuracy, by introducing waveform clustering, under the same depositional characteristics, deep network is used to predict reservoir parameters , to improve the prediction accuracy.
  • the flow of the main method is shown in Figure 1.
  • the first step is to establish a macro-geological feature zone.
  • the seismic attributes with high reliability are used as the division and basis for the classification of waveform features, and also provide the data basis for subsequent reservoir parameters; then, combined with the automatic clustering algorithm based on SOM unsupervised learning, the automatic division of waveform features is realized, and the waveform classification
  • the results are used as the basis for geological feature zoning to characterize different geological features; then, different reservoir parameter prediction models are constructed according to different geological features.
  • LSTM-RNN - Recurrent Neural Network
  • the method for predicting reservoir parameters based on geological feature constraints mainly includes the following steps.
  • the correlation between different types of seismic attributes of the target interval and the reservoir parameters is determined through intersection analysis.
  • the seismic attributes whose correlation exceeds the preset correlation threshold are selected as the dominant seismic attributes; then, for each dominant seismic attribute, the data of the seismic attribute is decomposed and reconstructed by singular spectrum analysis, wherein, In the reconstructed sequence, the sequence components whose contribution degree is higher than the preset contribution degree threshold are retained as the dominant component of the dominant seismic attribute according to the contribution degree of the seismic attribute.
  • the decomposed and reconstructed dominant seismic attributes will be used for waveform cluster analysis and processing in step S200.
  • the seismic data and logging data can be pre-processed first. For example, smooth the logging data so that the spectrum of the smoothed logging data matches the spectrum of the seismic data; normalize the matched seismic data and logging data; The bottom is the boundary, and the seismic data and logging data belonging to the target interval are intercepted from the normalized seismic data and logging data. Then, different deep neural network models are trained using the seismic data and logging data belonging to the target interval to optimize the model parameters of each deep neural network model.
  • Step S100 is a seismic attribute optimization technique based on intersection analysis and singular spectrum analysis.
  • the intersection analysis of different types of seismic attributes and reservoir parameters of the target interval is first carried out, and attributes with higher correlation or contribution are selected.
  • the optimized seismic attributes are represented as one-dimensional data, and a trajectory matrix is constructed.
  • the trajectory matrix is further decomposed and reconstructed, and the different components and different components of the attributes are rearranged according to their contribution degrees.
  • the dominant earthquake is determined.
  • the dominant component of the attribute provides a data basis for subsequent waveform classification and direct prediction of reservoir parameters.
  • the specific algorithm of singular spectrum analysis is as follows:
  • Embedding representing preferred attribute data as one-dimensional data: [x 1 , x 2 , ..., x N ]
  • N is the sequence length
  • the eigenvector U m corresponding to ⁇ m reflects the evolution of the time series.
  • X i represents the i-th column of the trajectory matrix X, Indicates the weight of the time evolution type reflected by Xi in the period of Xi +1 , xi +2 , ..., xi+L of the original sequence.
  • the signal is reconstructed through the time empirical orthogonal function and the time principal component.
  • the specific reconstruction process is as follows:
  • the reconstructed sequence is equal to the original sequence, namely:
  • the dominant attributes and the main components of the dominant attributes are obtained, which provides a data basis for subsequent reservoir prediction.
  • the SOM unsupervised clustering algorithm (Fig. 2) is used to realize the automatic division of waveform features, and the earthquake multi-attribute is selected as the input data, and the SOM unsupervised network training model and topology are designed. Output waveform classification results to provide constraint data for subsequent reservoir parameter prediction.
  • the specific algorithm of SOM is as follows:
  • Each node randomly initializes its own parameters.
  • the number of parameters for each node is the same as the dimension of the input data.
  • the construction of different deep neural networks with waveform features is further carried out. Since the seismic data has the characteristics of time series signals, and the logging data also has a certain correlation in the vertical direction, the LSTM-RNN (Long Short-Term Memory-Recurrent Neural Network) is preferred to construct a nonlinear multi-network prediction model under different geological characteristics (Fig. 3). ).
  • LSTM-RNN Long Short-Term Memory-Recurrent Neural Network
  • FIG. 3 The schematic diagram of the LSTM-RNN recurrent neural network model is shown in Figure 3.
  • An LSTM unit is composed of three threshold structures and one state vector transmission line.
  • the thresholds are the forget gate, the incoming gate, and the output gate.
  • the state vector transmission line is responsible for long-term memory. , only do some simple linear operations; 3 gates are responsible for the selection of short-term memory, and delete or add operations to the input vector through the threshold setting.
  • the forget gate ( Figure 4) is implemented by a sigmoid neural layer, whose role is to decide what information to let through the unit. 0 means "let no information through”, 1 means "let all information through”.
  • the role of the incoming gate ( Figure 5) is to decide how much new information to add to the cell state.
  • the implementation of the incoming gate requires two steps: first, a sigmod layer decides which information needs to be updated; a tanh layer generates an alternative to update content; in the next step, the two departments are combined by vector dot product to update the state of the unit.
  • the function of the output gate ( Figure 6) is to output the final result.
  • the implementation of the output gate requires two steps: first, a sigmoid layer is used to determine which part of the information will be output; then, the state vector is passed through a tanh layer, and then the tanh layer is passed. The output of the sigmoid layer is multiplied by the weight calculated by the sigmoid layer, so that the final output result is obtained.
  • i is the input gate
  • ⁇ is the logical sigmoid function
  • W xi , W hi , and W ci represent the weight matrix between the input feature vector, the hidden layer unit, the unit activation vector and the input gate, respectively
  • b i is the bias of the input gate.
  • f is the forget gate
  • W xf , W hf , W cf represent the weight matrix between the input feature vector, the hidden layer unit, the unit activation vector and the forget gate, respectively
  • b f is the bias of the forget gate
  • C is the Unit activation vector
  • W xc , W hc are the input feature vector, the weight matrix between the hidden layer unit and the unit activation vector, respectively, the weight matrix is a diagonal matrix
  • b c is the output gate offset value
  • o is the output Gate
  • W xo , W ho , W co represent the weight matrix between the input feature vector, hidden layer unit, unit activation vector and the output gate, respectively
  • b f is the bias of the forget gate
  • t as a subscript represents the sampling time
  • tanh is the activation function.
  • the input layer is the multi-attribute data next to the well
  • the output layer is the corresponding logging elastic parameters, such as the longitudinal wave velocity
  • the hidden layer is composed of: LSTM unit, full-connected layer, dropout layer, and regression layer, among which: LSTM unit is used for The time series features of logging data and seismic data are preserved; the full-connected layer is used as a classifier for the entire training network; the dropout layer is used to alleviate the occurrence of overfitting during network training and has a regularization effect; the regression layer is used as the network The output of the trained model.
  • w is the spatial variation coefficient
  • v 1 is the seismic trace of the neural network model that has been constructed
  • v 2 is the seismic trace of the neural network model to be constructed
  • w c represents the seismic trace of the neural network model that has been constructed and the seismic trace of the neural network model to be constructed.
  • the interpolation coefficient of the similarity of the trace waveform w d represents the interpolation coefficient of the distance between the seismic trace of the neural network model that has been constructed and the seismic trace of the neural network model to be constructed
  • c 12 represents the correlation between v 1 and v 2
  • d 12 represents v the distance between 1 and v2 , and represent the spatial positions of v 1 and v 2 , respectively
  • ⁇ is the adjustment factor
  • ⁇ c and ⁇ d are the exponential factors.
  • V p represents the reservoir parameters
  • f k (x 1 , x 2 ,...x N ) represents the neural network prediction model corresponding to the kth geological feature
  • wk , i, j represent the kth geological feature corresponding
  • the spatial variation coefficients of the neural network prediction model, x 1 , x 2 , ... x N represent different types of seismic attributes.
  • This area is a clastic rock reservoir type, and two sets of reservoirs are developed, namely O72 and O73 layers, O72 layer is turbidite channel sheet sandstone, the development of the whole area is stable, there are 3 effective logging wells in this area, respectively
  • PU_IA well, PU_IB well, PU_IC well Firstly, multiple attributes are optimized based on intersection analysis and singular spectrum analysis technology (Fig. 8). The preferred input attributes are: envelope attribute, relative wave impedance, and instantaneous amplitude attribute. Combined with SOM unsupervised clustering algorithm, waveform classification is performed (Fig. 9).
  • the waveforms are divided into three categories: weak energy, medium energy, and strong energy.
  • the three corresponding wells are PU_IC well, PU_IB well, and PU_IA well.
  • the deep neural network is trained by waveform features, and the preferred input data (Fig. 10) include: seismic trace, envelope property, thin layer factor, relative wave impedance, Hilbert property, instantaneous frequency, dominant frequency and the instantaneous phase attribute, the output data is: longitudinal wave velocity, a 5-layer deep neural network is designed (Fig. 11), the prediction model is constructed, the waveform classification result is used as the constraint data, and the multi-model elastic parameter prediction based on the deep recurrent neural network is finally realized.
  • the direct prediction technology of reservoir parameters based on LSTM-RNN cyclic neural network is developed, which realizes the direct prediction of reservoir parameters based on the set of spatially varying neural network prediction models, and effectively maintains the geological stratigraphic structure.
  • the prediction accuracy of reservoir parameters is further improved.
  • the present embodiment provides a reservoir parameter prediction device, which is characterized by comprising:
  • the attribute screening module is used to use the seismic data and logging data of the target interval to analyze the correlation between different types of seismic attributes of the target interval and the reservoir parameters. Select the dominant seismic attribute from the attributes;
  • the waveform classification module is used to classify the seismic waveforms of the target interval according to the waveform characteristics by using the preset waveform classification network model based on the dominant seismic attributes, and obtain the waveform classification results; wherein, different types of waveforms correspond to different geological features;
  • the model building module is used to construct different deep neural network models corresponding to different geological features with seismic attributes as input, reservoir parameters as output, and waveform classification results as constraints;
  • a model training module for using the seismic data and logging data of the target interval as training data and prediction data to train the different deep neural network models to optimize the model parameters of each deep neural network model;
  • a model fusion module used to fuse the trained different deep neural network models into a set of spatially varying neural network prediction models
  • the parameter prediction module is used to predict the reservoir parameters of the target interval by using a set of spatially varying neural network prediction models.
  • this embodiment provides a computer storage medium storing a computer program.
  • the computer storage medium when executed by one or more computer processors, implements the aforementioned method for predicting reservoir parameters.
  • the above-mentioned storage medium can be flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable memory
  • RAM random access memory
  • SRAM static random access memory
  • ROM read only memory
  • EEPROM programmable read-only memory
  • PROM programmable read-only memory
  • magnetic memory magnetic disk, optical disk, server, App (Application, application) application mall and so on.
  • this embodiment provides a computer device including a memory and a processor.
  • a computer program is stored in the memory, and when the computer program is executed by the processor, the aforementioned method for predicting reservoir parameters is executed.
  • the processor may be an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), or a Programmable Logic Device (Programmable Logic Device).
  • ASIC Application Specific Integrated Circuit
  • DSP Digital Signal Processor
  • DSPD Digital Signal Processing Device
  • PLD Programmable Logic Device
  • FPGA Field Programmable Gate Array
  • controller microcontroller
  • microprocessor or other electronic components.
  • the reservoir parameter prediction method described in any one of 5.
  • the memory can be implemented by any type of volatile or non-volatile storage device or their combination, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (Electrically Erasable) Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory ( Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM Static Random Access Memory
  • EEPROM Electrically Erasable Programmable Read-Only Memory
  • EPROM Erasable Programmable Read-Only Memory
  • PROM Programmable Read-Only Memory
  • ROM Read-Only Memory
  • magnetic memory flash memory
  • flash memory magnetic disk or optical disk.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more functions for implementing the specified logical function(s) executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures.
  • each functional module in each embodiment of the present disclosure may be integrated together to form an independent part, or each module may exist alone, or two or more modules may be integrated to form an independent part.
  • the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • the technical solutions of the present disclosure can be embodied in the form of software products in essence, or the parts that make contributions to the prior art or the parts of the technical solutions.
  • the computer software products are stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Remote Sensing (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Acoustics & Sound (AREA)
  • Geology (AREA)
  • Geophysics (AREA)
  • Environmental & Geological Engineering (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

一种基于地质特征约束的储层参数预测方法、装置和计算机存储介质以及计算机设备。根据目标层段的不同类型的地震属性与储层参数之间的相关度选出优势地震属性（S100）；基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果（S200）；以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型（S300）；将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合（S500）；利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测（S600）。

Description

基于地质特征约束的储层参数预测方法、装置和存储介质
本公开要求享有2020年9月7号提交的名称为“基于地质特征约束的储层参数预测方法、装置和存储介质”的中国专利申请CN 202010928912.0和名称为“储层参数预测方法、装置、存储介质及电子设备”的中国专利申请CN 202010931370.2的优先权，其全部内容通过引用并入本文中。
技术领域
本公开属于地球物理勘探技术领域，尤其涉及一种基于地质特征约束的储层参数预测方法、装置和计算机存储介质以及计算机设备。
背景技术
随着叠前深度偏移处理技术的发展及其在地震资料处理中日益广泛的应用，直接在深度域进行储层参数预测具有十分重要的意义。针对深度域地震储层预测技术，国内外主要的方法技术主要有三种：一是以Hampson-Russell软件公司部分模块、Geophysical Insight公司、BGP公司为代表的映射类方法，该技术主要通过神经网络建立多属性数据与测井数据的一个综合性的网络映射关系，进行储层参数直接预测，但是，在不同沉积环境下，用同一个模型预测可能会出现预测精度低的问题，同时，缺少地质特征约束，预测模型泛化能力差，预测结果容易过拟合，不符合宏观地质规律；二是以Paradigm公司、中石油西北分院、胜利油田为代表的基于深度域“子波”提取的反演类方法，该技术主要通过深度域“子波”提取，结合常规的反演技术，开展弹性参数直接预测，但是基于深度域数据的理论模型尚未建立，因此基础理论性不足；三是以Jason公司为代表的基于高精度速度体转换的深度域储层参数预测方法，该方法将深度域数据转化为时间域数据，进行常规储层参数预测，但是深时转化存在一定的转化累积误差，且费时费力，不利于提高储层预测精度。
发明内容
针对上述技术问题，本公开提出了一种基于地质特征约束的储层参数预测方法、装置和计算机存储介质以及计算机设备。本公开
根据本公开的第一方面，本公开提供的一种基于地质特征约束的储层参数预测方法，包括以下步骤：
S100、根据目标层段的不同类型的地震属性与储层参数之间的相关度，从不同类型的地震属性中选出优势地震属性；
S200、基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同类型的波形对应表征不同的地质特征；
S300、以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型；
S400、利用目标层段的地震数据和测井数据作为训练数据和预测数据，对所述不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数；
S500、将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合；
S600、利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
根据本公开的一个实施例，上述步骤S100包括以下步骤：
利用目标层段的地震数据和测井数据，通过交汇分析确定目标层段的不同类型的地震属性与储层参数之间的相关度，从不同类型的地震属性中选出相关度超过预设相关度阈值的地震属性，作为优势地震属性；
对于每一个优势地震属性，通过奇异谱分析对该地震属性的数据进行分解和重构，其中，在重构后的序列中按照对该地震属性的贡献度的大小保留贡献度高于预设贡献度阈值的序列分量，作为该优势地震属性的优势分量。
根据本公开的一个实施例，上述步骤S200中，所述波形分类网络模型为基于SOM无监督聚类算法设计的SOM无监督网络模型，该网络模型包括地震属性输入层和分类结果输出层。
根据本公开的一个实施例，所述地质特征包括沉积特征。
根据本公开的一个实施例，上述步骤S300中，每个所述深度神经网络模型 为LSTM-RNN循环神经网络模型，该网络模型包括地震属性输入层和储层参数输出层以及位于地震属性输入层与储层参数输出层之间的隐藏层；其中，所述隐藏层包括：
LSTM单元，用于保留地震数据和测井数据的时序特征；
full-connected层，用作训练网络模型的分类器
dropout层，用于缓解网络模型训练过程中的过拟合现象；
regression层，用作训练网络模型的输出。
根据本公开的一个实施例，所述步骤S400进一步包括：
对测井数据进行平滑处理，使得平滑处理后的测井数据的频谱与地震数据的频谱相互匹配；
对匹配后的地震数据和测井数据进行归一化处理；
以目的层段的顶底为边界，从归一化处理后的地震数据和测井数据中截取属于目的层段范围内的地震数据和测井数据；
利用属于目的层段范围内的地震数据和测井数据对不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数。
根据本公开的一个实施例，上述步骤S500中，基于波形的相似性和空间距离确定每个训练好的深度神经网络模型在空间变化神经网络预测模型集合中的空间变化系数。
根据本公开的一个实施例，上述步骤S600中，按照下式，将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合：
Figure PCTCN2021103487-appb-000001
式中：V p代表储层参数，f k(x 1，x 2，…x N)代表第k种地质特征下的深度神经网络模型，w k，i，j代表第k种地质特征下的深度神经网络模型的空间变化系数，x 1，x 2，…x N代表不同类型的地震属性。
根据本公开的一个实施例，按照下式，确定每个训练好的深度神经网络模型 在空间变化神经网络预测模型集合中的空间变化系数：
w＝λw c+(1-λ)w d
Figure PCTCN2021103487-appb-000002
Figure PCTCN2021103487-appb-000003
w c＝exp(-α cc 12 2)
w d＝exp(-α dd 12 2)
式中：w为空间变化系数，v 1为已构建深度神经网络模型的地震道，v 2为待构建深度神经网络模型的地震道，w c表示已构建深度神经网络模型地震道与待构建深度神经网络模型地震道波形相似性的插值系数，w d表示已构建深度神经网络模型地震道与待构建深度神经网络模型地震道距离的插值系数，c 12表示v 1和v 2之间的相关性，d 12表示表示v 1和v 2之间的距离，
Figure PCTCN2021103487-appb-000004
和
Figure PCTCN2021103487-appb-000005
分别表示v 1和v 2的空间位置，λ为调节因子，α c和α d为指数因子。
根据本公开的一个实施例，上述目的层段的储层参数包括目的层段的空间三维弹性参数体，所述方法还包括输出目的层段的空间三维弹性参数体的分布图。
根据本公开的第二方面，本公开还提供一种基于地质特征约束的储层参数预测装置，其特征在于，包括：
属性筛选模块，用于利用目标层段的地震数据和测井数据，分析目标层段的不同类型的地震属性与储层参数之间的相关度，根据所述相关度的大小从不同类型的地震属性中选出优势地震属性；
波形分类模块，用于基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同类型的波形对应表征不同的地质特征；
模型构建模块，用于以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型；
模型训练模块，用于利用目标层段的地震数据和测井数据作为训练数据和预测数据，对所述不同的深度神经网络模型进行训练，以优化每个深度神经网络模 型的模型参数；
模型融合模块，用于将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合；
参数预测模块，用于利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
根据本公开的第三方面，本公开还提供一种计算机存储介质，其特征在于，其中存储有可被处理器执行的计算机程序，该计算机程序在被处理器执行时实现上述基于地质特征约束的储层参数预测方法。
根据本公开的第四方面，本公开还提供一种计算机设备，其特征在于，包括存储器和处理器，所述处理器用于执行所述存储器中存储的计算机程序，以实现上述基于地质特征约束的储层参数预测方法。
与现有技术相比，本公开提供的融入内外部知识的暴力视频分类技术具有如下优点或有益效果：
1.本公开采用LSTM-RNN神经网络模型来描述地震属性与储层参数之间的非线性映射关系，既考虑了地震数据的上下关联性和测井数据的上下关联性，同时还兼顾了地震数据和测井数据的时序特征，建立了比现有技术更加准确的井震映射关系。
2.本公开通过引入波形聚类，在同一地质特征(例如，沉积特征)下对应一个深度网络模型，通过构建空间变化神经网络预测模型集合，以多类型地震属性数据为输入，以地质特征为约束，在不同的波形特征下使用不同的网络模型进行储层参数预测，有效提高了预测精度。
3.本公开是基于地质特征约束的非线性储层参数预测技术，能够实现对尤其是深度域储层参数的直接预测，提高储层参数预测精度及空间稳定性本公开，有助于进一步提高钻井成功率，降低油田勘探开发成本，提高油田生产效益。
本公开的其他特征和优点将在随后的说明书中阐述，并且，部分的从说明书中变得显而易见，或者通过实施本公开而了解。本公开的目的和其他优点可通过在说明书、权利要求书以及附图中所指出的结构来实现和获得。
附图说明
附图1是本公开实施例一的基于地质特征约束的储层参数预测方法流程示意 图；
附图2是本公开实施例一的储层参数预测方法中的SOM无监督聚类网络模型示意图；
附图3是本公开实施例一的储层参数预测方法中的LSTM-RNN循环神经网络模型示意图；
附图4是本公开实施例一的LSTM-RNN循环神经网络模型中的遗忘门示意图；
附图5是本公开实施例一的LSTM-RNN循环神经网络模型中的传入门示意图；
附图6是本公开实施例一的LSTM-RNN循环神经网络模型中的输出门示意图；
附图7是本公开实施例一的构建空间变化神经网络预测模型集合中的空间变化系数的示意图；
附图8是本公开实施例二的波形分类输入数据示意图；
附图9是本公开实施例二的波形分类结果沿O72切片示意图；
附图10是本公开实施例二的深度神经网络输入数据示意图；
附图11是本公开实施例二的深度神经网络模型构建示意图；
附图12(a)是本公开实施例二的基于深度神经网络的多模型弹性参数预测结果沿层切片示意图；
附图12(b)是本公开实施例二的基于深度神经网络的单模型弹性参数预测结果沿层切片示意图；
附图12(c)是本公开实施例二的时间域反演结果沿层切片示意图；
附图12(d)是本公开实施例二的工程开发部署方案示意图；
附图13(a)是本公开实施例二的PU_IA井原始测井曲线与多模型预测结果对比图；
附图13(b)是本公开实施例二的PU_IB井原始测井曲线与多模型预测结果对比图；
附图13(c)是本公开实施例二的PU_IC井原始测井曲线与多模型预测结果对比图；
附图13(d)是本公开实施例二的PU_IC井原始测井曲线与单模型预测结果 对比图。
具体实施方式
由于目前常见的深度域储层预测方法存在预测精度低、预测效率低以及不符合宏观地质特征等缺点，因此本公开提出了一种基于地质特征约束的深度域储层参数直接预测技术，以提高尤其深度域储层参数预测精度，提升空间预测稳定性，为后续钻井及油藏模拟提供合理的认识、高精度的数据，以支撑高效勘探开发。
本公开的核心思想是：针对在例如不同沉积环境下，使用同一个模型预测可能会导致预测精度骗低的问题，通过引入波形聚类，在同一沉积特征下，采用深度网络进行储层参数预测，以提高预测精度。主要方法的流程如图1所示，首先是建立宏观地质特征分带，通过计算深度域动力学、运动学、几何学等多类型的属性，并比对宏观地质背景及测井数据，优选相关性高的地震属性，作为波形特征分类的划分及依据，也作为后续储层参数提供数据基础；然后，结合基于SOM无监督学习的自动聚类算法，实现对波形特征的自动划分，以波形分类结果作为地质特征分带依据来表征不同的地质特征；接着，按照不同地质特征构建不同的储层参数预测模型，在不同的地质分带下，结合测井数据和地震属性数据，利用长短时记忆-循环神经网络(LSTM-RNN)构建不同井位置处的非线性网络预测模型，通过不断调参、优化模型参数，使单点预测模型泛化、收敛；最后，构建空间变化神经网络预测模型集合，以多类型地震属性数据为输入，以地质特征为约束，在不同的波形特征下使用不同的网络模型进行储层参数预测，最终获得三维空间参数体的综合预测结果。
为使本公开的目的，技术方案和优点更加清楚，以下结合实施例和附图对本公开作进一步的详细说明，借此对本公开如何应用技术手段解决技术问题，并达成技术效果的实现过程能充分理解并据以实施。
实施例一
如图1所示，本公开提供的基于地质特征约束的储层参数预测方法主要包括以下步骤。
S100、根据目标层段的不同类型的地震属性与储层参数之间的相关度，从不 同类型的地震属性中选出优势地震属性。
具体而言，首先利用目标层段的地震数据和测井数据，通过交汇分析确定目标层段的不同类型的地震属性与储层参数之间的相关度，根据所述相关度的大小从不同类型的地震属性中选出相关度超过预设相关度阈值的地震属性，作为优势地震属性；然后，对于每一个优势地震属性，通过奇异谱分析对该地震属性的数据进行分解和重构，其中，在重构后的序列中按照对该地震属性的贡献度的大小保留贡献度高于预设贡献度阈值的序列分量，作为该优势地震属性的优势分量。
经过分解和重构的优势地震属性将被用于步骤S200的波形聚类分析和处理。
S200、基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同的波形对应于不同的地质特征。
S300、以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型。
S400、利用目标层段的地震数据和测井数据作为训练数据和预测数据，对不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数。
在具体应用时，为了进一步提高预测结果的精确度，还可以首先对地震数据和测井数据做一些预处理。例如，对测井数据进行平滑处理，使得平滑处理后的测井数据的频谱与地震数据的频谱相互匹配；对匹配后的地震数据和测井数据进行归一化处理；以目的层段的顶底为边界，从归一化处理后的地震数据和测井数据中截取属于目的层段范围内的地震数据和测井数据。然后，再利用属于目的层段范围内的地震数据和测井数据对不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数。
S500、将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合；其中，利用波形相似性和空间距离确定每个训练好的深度神经网络模型在空间变化神经网络预测模型集合中的空间变化系数。
S600、利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
下面对各个步骤展开详细说明。
(1)步骤S100是基于交汇分析和奇异谱分析的地震属性优选技术。在该步 骤中，首先开展目标层段的不同类型的地震属性与储层参数的交汇分析，优选相关性或贡献度较高的属性。然后，将优选后的地震属性表示成一维数据，并构造出轨迹矩阵，进一步将轨迹矩阵进行分解、重构，按贡献度大小将属性的不同成分、不同分量进行重排，最终，确定优势地震属性的优势分量，为后续波形分类和储层参数直接预测提供数据基础。奇异谱分析的具体算法如下：
嵌入，将优选的属性数据表示为一维数据：[x 1，x 2，…，x N]
[x 1，x 2，…，x N]
其中，N为序列长度。
首先选择合适的窗口长度L，将原始时间序列进行滞后排列得到轨迹矩阵：
Figure PCTCN2021103487-appb-000006
通常情况下取L＜N/2，令K＝N-L+1，则轨迹矩阵X为L×K的矩阵：
Figure PCTCN2021103487-appb-000007
②分解，计算轨迹矩阵的协方差矩阵：
S＝X·X T   (3)
接下来，对S进行特征值分解得到λ 1＞λ 2＞…＞λ L≥0和对应的特征向量U 1，U 2，…，U L。此时，U＝[U 1，U 2，…，U L]，
Figure PCTCN2021103487-appb-000008
为原序列的奇异谱，并且有：
Figure PCTCN2021103487-appb-000009
其中：λ m对应的特征向量U m反映了时间序列的演变型。
③分组，假定所有的L个成分分为c个不相交的组，具体公式如下：
Figure PCTCN2021103487-appb-000010
④重构，首先计算迟滞序列X i在U m上的投影：
Figure PCTCN2021103487-appb-000011
其中，X i表示轨迹矩阵X的第i列，
Figure PCTCN2021103487-appb-000012
表示X i所反映的时间演变型在原序列的X i+1，x i+2，…，x i+L时段的权重。
接下来，通过时间经验正交函数和时间主成分来重构信号，具体重构过程如下：
Figure PCTCN2021103487-appb-000013
这样，重构序列的等于原序列，即：
Figure PCTCN2021103487-appb-000014
其中，
Figure PCTCN2021103487-appb-000015
为按主要性排序的第k个信号。
因此，通过交汇分析和奇异谱分析，获得优势属性及优势属性的主要分量，为后续储层预测提供数据基础。
(2)在多属性优选的基础上，通过SOM无监督聚类算法(图2)，来实现波形特征的自动划分，优选地震多属性为输入数据，设计SOM无监督网络训练模型及拓扑结构，输出波形分类结果，为后续储层参数预测提供约束数据。SOM具体算法如下：
①初始化：每个节点随机初始化自己的参数。每个节点的参数个数与输入数据的维度相同。
②对于每一个输入数据，找到与它最相配的节点。假设输入是D 维的数据，即X＝{x_i，i＝1，，D}，那么判别函数可以为欧几里得距离：
Figure PCTCN2021103487-appb-000016
③找到激活节点I(x)之后，我们也希望更新和它临近的节点。令S_ij表示节点i和j之间的距离，对于I(x)临近的节点，分配给它们一个更新权重：
Figure PCTCN2021103487-appb-000017
④接着就是更新节点的参数了。按照梯度下降法更新：
Δw ji＝η(t)*T j，I(x)(t)*(x i-w ji)   (12)
迭代，直至收敛。
因此，通过SOM无监督自动聚类技术，结合基于交汇分析和SSA奇异谱分析的多属性优选技术，实现波形特征自动划分。
(3)在波形分类的基础上，进一步开展分波形特征的不同深度神经网络构建。由于地震数据具有时序信号特征，且测井数据在纵向上也具有一定关联性，优选LSTM-RNN(长短时记忆-循环神经网络)进行不同地质特征下的非线性多网络预测模型构建(图3)。
LSTM-RNN循环神经网络模型的示意图为图3，一个LSTM单元是由3个门限结构和1个状态向量传输线组成的，门限分别是遗忘门，传入门，输出门；其中状态向量传输线负责长程记忆，只做一些简单的线性操作；3个门负责短时记忆的选择，通过门限设置对输入向量做删除或者添加操作。
遗忘门(图4)通过一个sigmoid神经层实现，它的作用是决定让哪些信息通过这个单元。0表示“不让任何信息通过”，1表示“让所有信息通过”。
传入门(图5)的作用是决定让多少新信息加入到单元状态中，传入门实现需要两个步骤：首先，一个sigmod层决定哪些信息需要更新；一个tanh层生成一个备选的用来更新内容；再下一步，将这两部门通 过向量点乘联合起来，对单元的状态进行一个更新。
输出门(图6)的作用是输出最后的结果，输出门实现需要两个步骤：首先，通过一个sigmoid层决定哪部分信息会被输出；接着，把状态向量通过一个tanh层，然后将tanh层的输出和sigmoid层计算出来的权重相乘，这样就得到了最后输出的结果。
LSTM单元的具体数学过程如下：
Figure PCTCN2021103487-appb-000018
其中：i为输入门，σ为逻辑sigmoid函数，W xi、W hi、W ci分别表示输入特征向量、隐藏层单元、单元激活向量与输入门之间的权重矩阵，b i为输入门的偏置量；f为遗忘门，W xf、W hf、W cf分别表示输入特征向量、隐藏层单元、单元激活向量与遗忘门之间的权重矩阵，b f为遗忘门的偏置量；C为单元激活向量，W xc、W hc分别为输入特征向量、隐藏层单元与单元激活向量之间的权重矩阵，所述权重矩阵为对角阵，b c为输出门的偏置值；o为输出门，W xo、W ho、W co分别表示输入特征向量、隐藏层单元、单元激活向量与输出门之间的权重矩阵，b f为遗忘门的偏置量；t作为下标时表示采样时刻，tanh为激活函数。
(4)训练基于LSTM-RNN的深度神经网络模型，并优化调整网络模型参数，使模型泛化、收敛。其输入层为井旁多属性数据，输出层为相应的测井弹性参数，如纵波速度，隐藏层则由：LSTM单元、full-connected层、dropout层、regression层组成，其中：LSTM单元用于保留测井数据和地震数据的时序特征；full-connected层作为整个训练网络的分类器；dropout层用于缓解网络训练过程中过拟合现象的发生，起到正则化的效果；regression层作为网络训练模型的输出。
(5)构建空间变化神经网络预测模型集合。在构建空间变化神经网络预测模型集合的基础上，利用波形相似性和空间距离构建每个神经网络模型的空间变化系数(图7)，实现空间变化神经网络预测模型集合的构建。空间变化系数的构建公式为：
w＝λw c+(1-λ)w d    (14)
Figure PCTCN2021103487-appb-000019
Figure PCTCN2021103487-appb-000020
w c＝exp(-α cc 12 2)     (17)
w d＝exp(-α dd 12 2)    (18)
式中：w为空间变化系数，v 1为已构建神经网络模型的地震道，v 2为待构建神经网络模型的地震道，w c表示已构建神经网络模型地震道与待构建神经网络模型地震道波形相似性的插值系数，w d表示已构建神经网络模型地震道与待构建神经网络模型地震道距离的插值系数，c 12表示v 1和v 2之间的相关性，d 12表示表示v 1和v 2之间的距离，
Figure PCTCN2021103487-appb-000021
和
Figure PCTCN2021103487-appb-000022
分别表示v 1和v 2的空间位置，λ为调节因子，α c和α d为指数因子。
最终，实现空变神经网络预测模型的构建。
(6)空间弹性参数预测。以多属性为输入数据，以弹性参数为输出数据，以波形分类结果为约束，通过空间变化神经网络预测模型集合实现空间弹性参数的预测。
Figure PCTCN2021103487-appb-000023
其中：V p代表储层参数，f k(x 1，x 2，…x N)代表第k种地质特征所对应的神经网络预测模型，w k，i，j代表第k种地质特征所对应的神经网络预测模型的空间变化 系数，x 1，x 2，…x N代表不同类型的地震属性。
实施例二
下面通过某工区的实际数据来说明本公开方法的有效性。该区为碎屑岩储层类型，共发育两套储层，分别为O72和O73层，O72层为浊积水道席状砂岩，全区发育稳定，该区有效测井共有3口，分别为PU_IA井、PU_IB井、PU_IC井。首先基于交汇分析和奇异谱分析技术优选多属性(图8)，优选输入属性分别为：包络属性、相对波阻抗、瞬时振幅属性，并结合SOM无监督聚类算法，进行波形分类(图9)，将波形共分为三类，为：能量较弱、能量中等、能量较强，分别对应的三口井为PU_IC井、PU_IB井、PU_IA井。在波形分类的基础上，分波形特征训练深度神经网络，优选输入数据(图10)包括：地震道、包络属性、薄层因子、相对波阻抗、希尔伯特属性、瞬时频率、主频及瞬时相位属性，输出数据为：纵波速度，设计5层深度神经网络(图11)，进行预测模型构建，将波形分类结果作为约束数据，最终实现基于深度循环神经网络的多模型弹性参数预测。
对比基于深度神经网络的多模型弹性参数预测结果与基于深度神经网络的单模型弹性参数预测结果、时间域反演结果以及工程开发部署方案(图12)，从平面趋势上可看出，本公开专利与基于单模型预测的纵波速度相比较，具有较好的空间预测精度，且较好地保留了整体宏观特征；本公开专利与时间域反演结果对比可看出，由于时间域反演结果受井及初始模型的限制，反演结果不能较好地表征宏观地质特征，综合图12-a、b、c、d分析，基于多模型的弹性参数直接预测技术具有较好的空间预测精度，且与开发部署方案具有较好的一致性。
对比PU_IA、PU_IB、PU_IC三口井的基于多模型预测结果与原始测井曲线(图13-a、b、c)，可见，纵波速度预测结果与原始测井曲线整体趋势一致，且与井吻合度均较高；进一步，对比PU_IC井的多模型预测结果与单模型预测结果，可见，基于多模型预测的纵波速度预测精度高于基于单模型预测的纵波速度。
综上所述，基于本公开专利研发了基于LSTM-RNN循环神经网络的储层参数直接预测技术，实现了基于空间变化神经网络预测模型集合的储层参数直接预测，有效保持了地质地层结构，进一步提高了储层参数预测精度。
实施例三
在上一实施例的基础上，本实施例提供一种储层参数预测装置，其特征在于，包括：
属性筛选模块，用于利用目标层段的地震数据和测井数据，分析目标层段的不同类型的地震属性与储层参数之间的相关度，根据所述相关度的大小从不同类型的地震属性中选出优势地震属性；
波形分类模块，用于基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同类型的波形对应表征不同的地质特征；
模型构建模块，用于以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型；
模型训练模块，用于利用目标层段的地震数据和测井数据作为训练数据和预测数据，对所述不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数；
模型融合模块，用于将训练好的所述不同的深度神经网络模型融合成空间变化神经网络预测模型集合；
参数预测模块，用于利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
实施例四
此外，本实施例提供一种计算机存储介质，所述计算机存储介质存储有计算机程序。
所述计算机存储介质被一个或多个计算机处理器执行时，实现如前所述的储层参数预测方法。
上述存储介质可以是闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘、服务器、App(Application，应用)应用 商城等等。
实施例五
另外，本实施例提供一种计算机设备，所述计算机设备包括存储器和处理器。
所述存储器上存储有计算机程序，所述计算机程序被所述处理器执行时，执行如前所述的储层参数预测方法。
处理器可以是专用集成电路(Application Specific Integrated Circuit，简称ASIC)、数字信号处理器(Digital Signal Processor，简称DSP)、数字信号处理设备(Digital Signal Processing Device，简称DSPD)、可编程逻辑器件(Programmable Logic Device，简称PLD)、现场可编程门阵列(Field Programmable Gate Array，简称FPGA)、控制器、微控制器、微处理器或其他电子元件实现，处理器可以用于执行上述实施例一至实施例五中任意一项所述的储层参数预测方法。
存储器可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，例如静态随机存取存储器(Static Random Access Memory，简称SRAM)，电可擦除可编程只读存储器(Electrically Erasable Programmable Read-Only Memory，简称EEPROM)，可擦除可编程只读存储器(Erasable Programmable Read-Only Memory，简称EPROM)，可编程只读存储器(Programmable Read-Only Memory，简称PROM)，只读存储器(Read-Only Memory，简称ROM)，磁存储器，快闪存储器，磁盘或光盘。
应该理解到，在上述实施例中所描述的装置和方法实施例仅仅是示意性的，例如，附图中的流程图和框图显示了根据本公开的多个实施例的装置、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或代码的一部分，所述模块、程序段或代码的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现方式中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，或者，也可以用 执行规定的功能或动作的专用的基于硬件的系统来实现，或者，还可以用专用硬件与计算机指令的组合来实现。
另外，在本公开各个实施例中的各功能模块可以集成在一起形成一个独立的部分，也可以是各个模块单独存在，也可以两个或两个以上模块集成形成一个独立的部分。
所述功能如果以软件功能模块的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。
还需要说明的是，以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步定义和解释。

Claims (13)

  1. 一种基于地质特征约束的储层参数预测方法，其特征在于，包括以下步骤：
    S100、根据目标层段的不同类型的地震属性与储层参数之间的相关度，从不同类型的地震属性中选出优势地震属性；
    S200、基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同类型的波形对应表征不同的地质特征；
    S300、以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型；
    S400、利用目标层段的地震数据和测井数据对所述不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数；
    S500、将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合；
    S600、利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
  2. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S100包括以下步骤：
    利用目标层段的地震数据和测井数据，通过交汇分析确定目标层段的不同类型的地震属性与储层参数之间的相关度，根据所述相关度的大小从不同类型的地震属性中选出相关度超过预设相关度阈值的地震属性，作为优势地震属性；
    对于每一个优势地震属性，通过奇异谱分析对该地震属性的数据进行分解和重构，其中，在重构后的序列中按照对该地震属性的贡献度的大小保留贡献度高于预设贡献度阈值的序列分量，作为该优势地震属性的优势分量。
  3. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S200中，所述波形分类网络模型为基于SOM无监督聚类算法设计的SOM无监督网络模型，该网络模型包括地震属性输入层和分类结果输出层。
  4. 根据权利要求1所述的储层参数预测方法，其特征在于，所述地质特征包括沉积特征。
  5. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S300中，每个所述深度神经网络模型为LSTM-RNN循环神经网络模型，该网络模型包括地震属性输入层和储层参数输出层以及位于地震属性输入层与储层参数输出层之间的隐藏层；其中，所述隐藏层包括：
    LSTM单元，用于保留地震数据和测井数据的时序特征；
    full-connected层，用作训练网络模型的分类器
    dropout层，用于缓解网络模型训练过程中的过拟合现象；
    regression层，用作训练网络模型的输出。
  6. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S400进一步包括：
    对测井数据进行平滑处理，使得平滑处理后的测井数据的频谱与地震数据的频谱相互匹配；
    对匹配后的地震数据和测井数据进行归一化处理；
    以目的层段的顶底为边界，从归一化处理后的地震数据和测井数据中截取属于目的层段范围内的地震数据和测井数据；
    利用属于目的层段范围内的地震数据和测井数据对不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数。
  7. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S500中，基于波形的相似性和空间距离确定每个训练好的深度神经网络模型在空间变化神经网络预测模型集合中的空间变化系数。
  8. 根据权利要求1所述的储层参数预测方法，其特征在于，所述步骤S600中，按照下式，将训练好的不同的深度神经网络模型融合成空间变化神经网络预 测模型集合：
    Figure PCTCN2021103487-appb-100001
    其中：V p代表储层参数，f k(x 1，x 2，…x N)代表第k种地质特征所对应的深度神经网络模型，w k，i，j代表第k种地质特征所对应的深度神经网络模型的空间变化系数，x 1，x 2，…x N代表不同类型的地震属性。
  9. 根据权利要求5所述的储层参数预测方法，其特征在于，按照下式，确定每个训练好的深度神经网络模型在空间变化神经网络预测模型集合中的空间变化系数：
    w＝λw c+(1-λ)w d
    Figure PCTCN2021103487-appb-100002
    Figure PCTCN2021103487-appb-100003
    w c＝exp(-α cc 12 2)
    w d＝exp(-α dd 12 2)
    式中：w为空间变化系数，v 1为已构建深度神经网络模型的地震道，v 2为待构建深度神经网络模型的地震道，w c表示已构建深度神经网络模型地震道与待构建深度神经网络模型地震道波形相似性的插值系数，w d表示已构建深度神经网络模型地震道与待构建深度神经网络模型地震道距离的插值系数，c 12表示v 1和v 2之间的相关性，d 12表示表示v 1和v 2之间的距离，
    Figure PCTCN2021103487-appb-100004
    和
    Figure PCTCN2021103487-appb-100005
    分别表示v 1和v 2的空间位置，λ为调节因子，α c和α d为指数因子。
  10. 根据权利要求1所述的储层参数预测方法，其特征在于，所述目的层段的储层参数包括目的层段的空间三维弹性参数体，所述方法还包括输出目的层段的空间三维弹性参数体的分布图。
  11. 一种基于地质特征约束的储层参数预测装置，其特征在于，包括：
    属性筛选模块，用于利用目标层段的地震数据和测井数据，分析目标层段的不同类型的地震属性与储层参数之间的相关度，根据所述相关度的大小从不同类型的地震属性中选出优势地震属性；
    波形分类模块，用于基于优势地震属性，利用预设的波形分类网络模型将目标层段的地震波形按照波形特征进行分类，获得波形分类结果；其中，不同类型的波形对应表征不同的地质特征；
    模型构建模块，用于以地震属性为输入，以储层参数为输出，以波形分类结果为约束，构建不同的地质特征所对应的不同的深度神经网络模型；
    模型训练模块，用于利用目标层段的地震数据和测井数据对所述不同的深度神经网络模型进行训练，以优化每个深度神经网络模型的模型参数；
    模型融合模块，用于将训练好的不同的深度神经网络模型融合成空间变化神经网络预测模型集合；
    参数预测模块，用于利用空间变化神经网络预测模型集合对目标层段的储层参数进行预测。
  12. 一种计算机存储介质，其特征在于，其中存储有可被处理器执行的计算机程序，该计算机程序在被处理器执行时实现上述权利要求1至10中任意一项所述基于地质特征约束的储层参数预测方法。
  13. 一种计算机设备，其特征在于，包括存储器和处理器，所述处理器用于执行所述存储器中存储的计算机程序，所述计算机程序用于实现上述权利要求1至10中任意一项所述基于地质特征约束的储层参数预测方法。
PCT/CN2021/103487 2020-09-07 2021-06-30 基于地质特征约束的储层参数预测方法、装置和存储介质 Ceased WO2022048288A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/005,011 US20230314649A1 (en) 2020-09-07 2021-06-30 Reservoir parameter prediction method and apparatus based on geological characteristic constraint, and storage medium

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202010928912.0A CN114152977B (zh) 2020-09-07 2020-09-07 基于地质特征约束的储层参数预测方法、装置和存储介质
CN202010928912.0 2020-09-07
CN202010931370.2 2020-09-07
CN202010931370.2A CN114152978B (zh) 2020-09-07 2020-09-07 储层参数预测方法、装置、存储介质及电子设备

Publications (1)

Publication Number Publication Date
WO2022048288A1 true WO2022048288A1 (zh) 2022-03-10

Family

ID=80491589

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/103487 Ceased WO2022048288A1 (zh) 2020-09-07 2021-06-30 基于地质特征约束的储层参数预测方法、装置和存储介质

Country Status (2)

Country Link
US (1) US20230314649A1 (zh)
WO (1) WO2022048288A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114609668A (zh) * 2022-03-11 2022-06-10 西安交通大学 一种基于散射变换和神经网络的优质储层识别方法、装置、设备及存储介质
CN115267883A (zh) * 2022-06-23 2022-11-01 华南理工大学 地震响应预测模型训练及预测方法、系统、设备及介质
CN115373028A (zh) * 2022-04-12 2022-11-22 中国石油大学(华东) 一种碳酸盐岩缝洞型油藏放空漏失井段测井曲线重构方法
CN115826039A (zh) * 2022-11-09 2023-03-21 中国地质科学院 一种时间切片分类模型训练方法、系统及应用方法、系统
CN115936234A (zh) * 2022-12-21 2023-04-07 成都理工大学 一种基于深度学习的薄储层空间分布预测方法
CN116089076A (zh) * 2022-12-30 2023-05-09 清华大学 数据处理方法、装置、计算机设备和存储介质
CN116186826A (zh) * 2022-11-29 2023-05-30 清华大学 基于数据-力学耦合驱动图神经网络的隔震支座设计方法
CN116578871A (zh) * 2023-05-19 2023-08-11 西南石油大学 一种基于聚类匹配思想的措施增油量预测方法
CN117250659A (zh) * 2022-06-09 2023-12-19 中国石油化工股份有限公司 基于断点剔除的地层超剥点识别方法

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116976221B (zh) * 2023-08-10 2024-05-17 西安理工大学 基于冲蚀特性的堰塞体溃决峰值流量预测方法及存储介质
US12320944B1 (en) 2023-12-05 2025-06-03 China Petroleum & Chemical Corporation Method for calibrating prestack seismic inversion using fully connected neural networks
CN118090107A (zh) * 2024-04-24 2024-05-28 华侨大学 一种考虑地震动影响的多结构地震反应智能预测方法
CN118884526B (zh) * 2024-07-12 2025-03-07 中国地震局地质研究所 一种重复地震识别方法、装置、介质及设备
CN119001849B (zh) * 2024-08-21 2025-08-05 西南石油大学 一种基于属性聚类的低速泥岩识别方法和系统
CN118885724B (zh) * 2024-09-27 2025-01-24 上海水伊源科技有限公司 一种基于光量子分解技术的地表水污染处理系统及方法
CN119416026B (zh) * 2024-10-30 2026-02-03 浙江爱达科技有限公司 一种基于改进SSA与DMSANet的心电图分类方法
CN120254967B (zh) * 2025-05-30 2026-04-03 东北石油大学三亚海洋油气研究院 一种页岩油产能预测方法、装置、介质和设备
CN120524446B (zh) * 2025-07-25 2025-09-30 西南石油大学 基于双重注意力卷积lstm的随钻测井储层参数动态预测方法
CN121069476B (zh) * 2025-09-15 2026-05-12 长江大学 一种基于人工智能识别地质包络体中层位边界的方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102650702A (zh) * 2012-05-03 2012-08-29 中国石油天然气股份有限公司 一种地震波形分析及储层预测方法和装置
CN103487832A (zh) * 2013-09-12 2014-01-01 电子科技大学 一种三维地震信号中的有监督波形分类方法
CN111025384A (zh) * 2019-11-26 2020-04-17 中国石油天然气股份有限公司 基于波形分类交汇融合的储层预测方法和装置
US20200160173A1 (en) * 2017-07-21 2020-05-21 Landmark Graphics Corporation Deep Learning Based Reservoir Modeling
CN111239824A (zh) * 2020-04-03 2020-06-05 辽宁工程技术大学 一种火山岩岩相的预测方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018184193A1 (en) * 2017-04-07 2018-10-11 Intel Corporation Advanced artificial intelligence agent for modeling physical interactions
ES2930550T3 (es) * 2017-04-07 2022-12-16 Intel Corp Métodos y aparatos para canalización de ejecución de red de aprendizaje profundo en plataforma multiprocesador
US11263490B2 (en) * 2017-04-07 2022-03-01 Intel Corporation Methods and systems for budgeted and simplified training of deep neural networks
US11790223B2 (en) * 2017-04-07 2023-10-17 Intel Corporation Methods and systems for boosting deep neural networks for deep learning
US10726514B2 (en) * 2017-04-28 2020-07-28 Intel Corporation Compute optimizations for low precision machine learning operations
US11846748B2 (en) * 2019-12-16 2023-12-19 Landmark Graphics Corporation, Inc. Deep learning seismic attribute fault predictions
CN111488051A (zh) * 2020-03-06 2020-08-04 复旦大学 基于cpu和fpga协同计算的云端深度神经网络优化方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102650702A (zh) * 2012-05-03 2012-08-29 中国石油天然气股份有限公司 一种地震波形分析及储层预测方法和装置
CN103487832A (zh) * 2013-09-12 2014-01-01 电子科技大学 一种三维地震信号中的有监督波形分类方法
US20200160173A1 (en) * 2017-07-21 2020-05-21 Landmark Graphics Corporation Deep Learning Based Reservoir Modeling
CN111025384A (zh) * 2019-11-26 2020-04-17 中国石油天然气股份有限公司 基于波形分类交汇融合的储层预测方法和装置
CN111239824A (zh) * 2020-04-03 2020-06-05 辽宁工程技术大学 一种火山岩岩相的预测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
GAO GAI;HUANG LI-GANG;XU LEI-MING;HUANG YAN: "Application of Neural Networks Multi-attribute Reservoir Prediction Technology in MZ Area", PROCEEDINGS OF INTERNATIONAL FIELD EXPLORATION AND DEVELOPMENT CONFERENCE 2019, 16 October 2019 (2019-10-16), pages 3487 - 3492, XP055907671, DOI: 10.26914/c.cnkihy.2019.050600 *
ZHAO ZHONGQUAN;HE ZHENHUA;WAN XIAOMING;SHUAI QINGWEI: "A Study on Fluid Prediction Based on the Classification of Sensitive Seismic Attributes", JOURNAL OF SOUTHWEST PETROLEUM UNIVERSITY(SCIENCE & TECHNOLOGY EDITION), vol. 38, no. 3, 1 June 2016 (2016-06-01), pages 75 - 81, XP055907665, ISSN: 1674-5086, DOI: 10.11885/j.issn.1674-5086.2014.04.13.03 *

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114609668B (zh) * 2022-03-11 2023-09-19 西安交通大学 一种基于散射变换和神经网络的优质储层识别方法、装置、设备及存储介质
CN114609668A (zh) * 2022-03-11 2022-06-10 西安交通大学 一种基于散射变换和神经网络的优质储层识别方法、装置、设备及存储介质
CN115373028A (zh) * 2022-04-12 2022-11-22 中国石油大学(华东) 一种碳酸盐岩缝洞型油藏放空漏失井段测井曲线重构方法
CN117250659A (zh) * 2022-06-09 2023-12-19 中国石油化工股份有限公司 基于断点剔除的地层超剥点识别方法
CN115267883B (zh) * 2022-06-23 2023-08-29 华南理工大学 地震响应预测模型训练及预测方法、系统、设备及介质
CN115267883A (zh) * 2022-06-23 2022-11-01 华南理工大学 地震响应预测模型训练及预测方法、系统、设备及介质
CN115826039A (zh) * 2022-11-09 2023-03-21 中国地质科学院 一种时间切片分类模型训练方法、系统及应用方法、系统
CN115826039B (zh) * 2022-11-09 2023-07-14 中国地质科学院 一种时间切片分类模型训练方法、系统及应用方法、系统
CN116186826A (zh) * 2022-11-29 2023-05-30 清华大学 基于数据-力学耦合驱动图神经网络的隔震支座设计方法
CN116186826B (zh) * 2022-11-29 2023-08-25 清华大学 基于数据-力学耦合驱动图神经网络的隔震支座设计方法
CN115936234A (zh) * 2022-12-21 2023-04-07 成都理工大学 一种基于深度学习的薄储层空间分布预测方法
CN116089076A (zh) * 2022-12-30 2023-05-09 清华大学 数据处理方法、装置、计算机设备和存储介质
CN116089076B (zh) * 2022-12-30 2023-11-28 清华大学 数据处理方法、装置、计算机设备和存储介质
CN116578871A (zh) * 2023-05-19 2023-08-11 西南石油大学 一种基于聚类匹配思想的措施增油量预测方法

Also Published As

Publication number Publication date
US20230314649A1 (en) 2023-10-05

Similar Documents

Publication Publication Date Title
US20230314649A1 (en) Reservoir parameter prediction method and apparatus based on geological characteristic constraint, and storage medium
CN114152977B (zh) 基于地质特征约束的储层参数预测方法、装置和存储介质
CN116047583B (zh) 基于深度卷积神经网络的自适应波阻抗反演方法及系统
Aleardi et al. 1D elastic full‐waveform inversion and uncertainty estimation by means of a hybrid genetic algorithm–Gibbs sampler approach
CN107688201B (zh) 基于rbm地震叠前信号聚类方法
CN112965103B (zh) 一种多重孔隙储层叠前地震概率化多道反演方法
CN111596978A (zh) 用人工智能进行岩相分类的网页显示方法、模块和系统
Zhao et al. Rock-physics-guided machine learning for shear sonic log prediction
Song et al. Gas-bearing prediction of tight sandstone reservoir using semi-supervised learning and transfer learning
Chen et al. Prediction of shear wave velocity based on a hybrid network of two‐dimensional convolutional neural network and gated recurrent unit
Alameedy et al. Evaluating machine learning techniques for carbonate formation permeability prediction using well log data
CN112147679A (zh) 在模糊逻辑框架下基于弹性参数的岩性预测方法和装置
Bougher Machine learning applications to geophysical data analysis
Chopra et al. Unsupervised machine learning applications for seismic facies classification
CN115880455A (zh) 基于深度学习的三维智能插值方法
CN115032691B (zh) 一种针对钻井土层中薄互层砂地比概率化地质分析方法
CN115936234A (zh) 一种基于深度学习的薄储层空间分布预测方法
CN111273346A (zh) 去除沉积背景的方法、装置、计算机设备及可读存储介质
CN112882123B (zh) 一种基于两步法的cnn井震联合反演方法、系统及应用
Xu et al. Pre-Stack Seismic Inversion via Global Optimization with an Accurate Hessian Matrix and the Three-Variable Cauchy Distribution for Exact Zoeppritz Equations
CN110488350A (zh) 基于卷积神经网络的地震反演大数据生成方法
CN115808710A (zh) 一种致密储层叠前地震数据特征提取方法
RU2836749C1 (ru) Способ и устройство прогнозирования параметров резервуара, основанные на ограничении геологических характеристик, и среда хранения данных
Mao et al. P-Impedance and V p/V s prediction based on AVO inversion scheme with deep feedforward neural network: a case study from tight sandstone reservoir
Malvić et al. Neural networks in petroleum geology as interpretation tools

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21863356

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2023100090

Country of ref document: RU

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21863356

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 523442220

Country of ref document: SA

WWE Wipo information: entry into national phase

Ref document number: 523442220

Country of ref document: SA

WWG Wipo information: grant in national office

Ref document number: 523442220

Country of ref document: SA

WWG Wipo information: grant in national office

Ref document number: 2023100090

Country of ref document: RU