CN112906760B - Horizontal well fracturing segment segmentation method, system, equipment and storage medium - Google Patents
Horizontal well fracturing segment segmentation method, system, equipment and storage medium Download PDFInfo
- Publication number
- CN112906760B CN112906760B CN202110130584.4A CN202110130584A CN112906760B CN 112906760 B CN112906760 B CN 112906760B CN 202110130584 A CN202110130584 A CN 202110130584A CN 112906760 B CN112906760 B CN 112906760B
- Authority
- CN
- China
- Prior art keywords
- horizontal well
- graph
- neural network
- matrix
- segmentation
- 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.)
- Active
Links
- 230000011218 segmentation Effects 0.000 title claims abstract description 113
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000003860 storage Methods 0.000 title claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims abstract description 92
- 239000011159 matrix material Substances 0.000 claims abstract description 89
- 238000004364 calculation method Methods 0.000 claims abstract description 45
- 239000013598 vector Substances 0.000 claims abstract description 25
- 238000010586 diagram Methods 0.000 claims abstract description 24
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 16
- 238000005457 optimization Methods 0.000 claims abstract description 13
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims abstract description 10
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000001228 spectrum Methods 0.000 claims abstract description 10
- 239000010410 layer Substances 0.000 claims description 49
- 238000012549 training Methods 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 18
- 238000010276 construction Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 5
- 239000002356 single layer Substances 0.000 claims description 4
- 239000002131 composite material Substances 0.000 description 9
- 238000011161 development Methods 0.000 description 9
- 230000018109 developmental process Effects 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 9
- 208000010392 Bone Fractures Diseases 0.000 description 6
- 206010017076 Fracture Diseases 0.000 description 6
- 238000013507 mapping Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 4
- 238000005086 pumping Methods 0.000 description 4
- 239000011435 rock Substances 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000004215 Carbon black (E152) Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 229930195733 hydrocarbon Natural products 0.000 description 1
- 150000002430 hydrocarbons Chemical class 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH OR ROCK DRILLING; MINING
- E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B43/00—Methods or apparatus for obtaining oil, gas, water, soluble or meltable materials or a slurry of minerals from wells
- E21B43/25—Methods for stimulating production
- E21B43/26—Methods for stimulating production by forming crevices or fractures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Mining & Mineral Resources (AREA)
- Geology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Environmental & Geological Engineering (AREA)
- Fluid Mechanics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geochemistry & Mineralogy (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a method, a system, equipment and a storage medium for segmenting a fracturing section of a horizontal well, which are used for initializing weights of a graph neural network and weights of all connection layers; constructing an input diagram according to the geological information of the horizontal well; creating a self-connection weighted adjacency matrix of the input graph, and carrying out Laplace spectrum decomposition to obtain a Laplace matrix; solving a feature vector matrix of the Laplace matrix, and inputting the feature vector matrix into a graph neural network for forward propagation calculation to obtain a network prediction segmentation scheme; calculating a difference value between a network predicted segmentation scheme and a manual segmentation scheme by using a cross entropy loss function, and back-propagating an update weight by using an Adam optimization algorithm; until all the geological information of the horizontal well is completely trained, obtaining a final trained graphic neural network; and inputting the geological information of the horizontal well to be segmented into the graph neural network to obtain a horizontal well fracturing segment segmentation scheme. The segmentation efficiency is greatly improved, and the segmentation quality is kept stable.
Description
Technical Field
The invention belongs to the field of unconventional oil and gas exploitation, and relates to a horizontal well fracturing section segmentation method, a horizontal well fracturing section segmentation system, a horizontal well fracturing section segmentation equipment and a horizontal well fracturing section segmentation storage medium.
Background
Along with the development of shale gas and compact oil gas in China, unconventional oil gas resources have become a new development hot spot in recent years. At present, the staged multi-cluster fracturing technology of the horizontal well is one of core technologies for unconventional oil and gas resource development, and in the horizontal well construction, a staged fracturing development mode of cluster perforation-composite bridge plug combination is widely applied. Compared with other development modes, the method has the advantages of realizing large-displacement injection, clustered perforation, staged volume fracturing, high operation efficiency and the like. Through dividing the horizontal well into a plurality of fracturing sections, carrying out multi-cluster perforation in the sections, a plurality of hydraulic cracks can be formed simultaneously under single pump injection, and construction cost is effectively reduced.
The staged fracturing technology of the clustered perforation-composite bridge plug needs to divide a horizontal well section into a plurality of sections (the control distance of one section is 100-150 m), wherein the first section adopts an oil pipe, a continuous oil pipe and a cable crawler to carry out perforation and then fracturing, and the other sections adopt the clustered perforation-composite bridge plug linkage process technology for construction. The combined instrument string is put into the well by a cable and is pushed in a pumping mode of a high-inclination horizontal well Duan Yongshui, namely the hydraulic pumping technology. Firstly, a previous section is plugged by a composite bridge plug, then the section is subjected to shower hole splitting, a combined instrument string is started, and then the section is subjected to volume fracturing construction.
The first step of the staged fracturing technology of the clustered perforation-composite bridge plug is horizontal well staging, firstly, geological investigation is conducted on a logging to obtain geological information such as gas measurement values, porosity, rock stratum distribution, predicted fracture sections and the like of the horizontal well, and then according to the geological information, horizontal well sections which are located in the same small layer, have natural fracture development sections, have similar reservoir parameters, have similar rock mechanical parameters and the like and have similar geological structures are divided into the same stage of perforation fracturing sections, so that the horizontal well is divided into a plurality of fracturing sections. The quality of the level of the section directly determines perforation quality, the effectiveness degree of hydraulic fracture and the effect of volume fracturing construction, and the perforation quality, the effectiveness degree of the hydraulic fracture and the effect of volume fracturing construction directly influence the oil gas output efficiency and the output quantity, so that the level well section is a key link of the sectional fracturing technology of the clustered perforation-composite bridge plug.
At present, the fracturing section of the horizontal well is segmented by adopting a manual segmentation mode, the horizontal well is segmented according to geological information by manpower, but the manual segmentation efficiency is low, the labor cost is too high, the quality of the manual segmentation is unstable, some production logging data show that some fracturing sections are not reasonably divided, part of perforation clusters cannot form effective hydraulic cracks, part of fracturing sections cannot form effective fracturing, so that the oil gas output efficiency is reduced, and the waste of oil gas resources is caused. Therefore, how to improve the segmentation efficiency, reduce the segmentation cost and improve the segmentation quality becomes a problem to be solved urgently.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a horizontal well fracturing segment segmentation method, a horizontal well fracturing segment segmentation system, a horizontal well fracturing segment segmentation device and a horizontal well fracturing segment storage medium, wherein the segmentation quality is kept close to that of a manual segmentation, and the horizontal well fracturing segment segmentation method, the horizontal well fracturing segment storage device and the horizontal well fracturing segment storage medium are efficient, rapid and automatic in segmentation, so that the segmentation efficiency is greatly improved, the labor cost is reduced, and the segmentation quality is kept stable.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
A horizontal well fracturing section segmentation method comprises the following steps of;
Step 1, initializing weights of a graph neural network and weights of a full-connection layer;
Step 2, constructing an input diagram according to the geological information of the horizontal well;
step 3, creating a self-connection weighted adjacent matrix of the input diagram, and carrying out Laplacian spectrum decomposition on the self-connection weighted adjacent matrix to obtain a Laplacian matrix;
Step 4, obtaining a feature vector matrix of the Laplace matrix, and inputting the feature vector matrix into a graph neural network to perform forward propagation calculation to obtain a network prediction segmentation scheme;
step 5, acquiring a manual segmentation scheme of the horizontal well fracturing section predicted by the network, calculating a difference value between the segmentation scheme predicted by the network and the manual segmentation scheme by using a cross entropy loss function, and back-propagating an update weight by using an Adam optimization algorithm;
Step 6, circulating the steps 2 to 5 until all the geological information of the horizontal well is completely trained, and obtaining a final trained graphic neural network;
and 7, inputting geological information of the horizontal well to be segmented into the graphic neural network to obtain a horizontal well fracturing segment segmentation scheme.
Preferably, in step 2, the specific process of constructing the input graph is as follows: determining the minimum sectional length of the horizontal well and the total length of the horizontal well, enabling the minimum sectional length to be the initial value of the graph nodes, enabling each section to be an input graph node, and calculating the number of the graph nodes; according to the horizontal well geological information of the segments represented by the graph nodes and the distance relation between each segment, weighted edges between the graph nodes are created to represent weights of the two graph nodes and edges between the graph nodes, and all graph nodes and all weighted edges are assembled to form an input graph of the graph neural network.
Further, the calculation process of the number of the graph nodes is as follows:
Where total_length represents the total length of the horizontal well, min_length represents the minimum segment length, and X num represents the number of nodes of the graph.
Preferably, in step 3, elements of the ith row and the jth column in the matrix of the self-connected weighted adjacency matrixThe values of (2) are:
wherein X (i, j) represents the weighted edge weight between the ith node and the jth node obtained in the step 3;
The laplace matrix is:
In the middle of Self-connecting weighted adjacency matrix for the graph obtained in step 4,/>For/>Degree matrix of L is pair/>And carrying out Laplace spectrum decomposition and normalization to obtain the Laplace matrix.
Preferably, the specific process of the step 4 is that a feature vector matrix of the Laplace matrix is obtained, a first layer feature vector matrix of the hidden state of the input diagram and a normalized Laplace matrix are input to a first layer in the diagram neural network for calculation, and a calculation result is obtained; performing nonlinear activation processing on the calculation result by using Relu activation functions to finish calculation of the single-layer graph neural network until the feature vector matrix of each layer of the graph neural network is updated; and inputting all updated results of the graph neural network into a full-connection layer to classify nodes, wherein the nodes classified into one type are used as one section, so that a horizontal well segmentation scheme predicted by the graph neural network is obtained.
Preferably, in step 5, the cross entropy loss function is:
Wherein Y represents a manual segmentation scheme, Segmentation scheme representing network predictions, Y i and/>Representing each element in the two vectors separately;
the Adam optimization algorithm comprises the following calculation processes:
t←t+1
mt←β1·mt-1+(1-β1)·gt
Wherein t represents a time step t; f (θ) represents the objective function to be optimized, in the present invention the cross entropy loss function; θ represents a network weight parameter to be updated, θ t-1 represents a parameter of a previous time step, and θ t represents a parameter of a current time step updated after one iteration; beta 1 and beta 2 are two important super parameters of the algorithm, and the general values are 0.9 and 0.999 respectively; alpha is learning rate, generally the initial value is 0.01 or 0.001, and learning rate attenuation is carried out after 10-20 iterations; epsilon represents the optimization algorithm bias and generally takes on a value of 10 -8.
A horizontal well fracturing segment segmentation system comprising:
the weight initialization module is used for initializing the weights of the graphic neural network and the weights of the full-connection layer;
The input diagram construction module is used for constructing an input diagram according to the geological information of the horizontal well;
The Laplace matrix calculation module is used for creating a self-connection weighted adjacent matrix of the input graph, and carrying out Laplace spectrum decomposition on the self-connection weighted adjacent matrix to obtain a Laplace matrix;
the network prediction segmentation scheme calculation module is used for solving a feature vector matrix of the Laplace matrix, inputting the feature vector matrix into the graph neural network for forward propagation calculation, and obtaining a network prediction segmentation scheme;
the backward propagation updating weight module is used for acquiring an artificial segmentation scheme of the horizontal well fracturing section predicted by the network, calculating a difference value between the segmentation scheme predicted by the network and the artificial segmentation scheme by using a cross entropy loss function, and then using an Adam optimization algorithm to backward propagate updating weight;
The graph neural network training completion acquisition module is used for repeatedly inputting the calculation processes of the graph construction module, the Laplace matrix calculation module, the network prediction segmentation scheme calculation module and the back propagation updating weight module until all the horizontal well geological information is completely trained, and obtaining a graph neural network with the final training completed;
the horizontal well fracturing segment segmentation module is used for inputting the geological information of the horizontal well to be segmented into the graphic neural network to obtain a horizontal well fracturing segment segmentation scheme.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the steps of the horizontal well fracturing segment segmentation method of any one of the preceding claims.
A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the horizontal well fracturing segment segmentation method according to any of the preceding claims.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, according to the non-Euclidean space characteristics of logging geological information, the data modeling of the graph form is carried out on the logging geological data and is used as a graph neural network training set, and according to the strong capability of the graph neural network in the aspect of processing the characteristics of the non-Euclidean space data, the graph neural network is constructed, and the existing manual segmentation scheme is utilized to enable the graph neural network to learn the relation between the geological data and the segmentation scheme. After the graphic neural network is trained, the horizontal well which is not subjected to manual segmentation is automatically segmented without manual intervention by using the network, so that the segmentation quality can be effectively and rapidly automatically segmented under the condition that the segmentation quality is kept close to that of the manual segmentation, the segmentation efficiency is greatly improved, the labor cost is reduced, and the segmentation quality is kept stable.
Drawings
FIG. 1 is a schematic diagram of the neural network used in the present invention;
FIG. 2 is a schematic illustration of a horizontal well section configuration in which the present invention is to be practiced;
FIG. 3 is a schematic diagram of the neural network training process provided by the present invention;
FIG. 4 is a schematic diagram of the process of automatically segmenting the trained neural network according to the present invention;
fig. 5 is a flow chart of an implementation of the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
FIG. 1 is a schematic diagram of the neural network for calculation according to the present invention. The image neural network is mainly divided into two parts of a hidden layer and an output layer, wherein the hidden layer is used for carrying out matrix multiplication operation on an output hidden state H l of the upper layer, a feature vector normalized Laplacian matrix L of an image node and a weight W l of the layer, and then carrying out nonlinear activation processing on a result by using a relu function so as to update the hidden state;
Hl+1＝σ(LHlWl)
The output layer is used for receiving the final calculation result output by the hidden layer, namely the final updated graph information, and then calculating the graph information to obtain the final calculation result. The form of the output layer and the meaning of the calculation result depend on specific downstream tasks, and in the invention, the output layer is a fully-connected layer, classifying operation is carried out on the graph information, and the final output data represents a horizontal well segmentation scheme.
In the middle ofFor the final calculation result of the neural network of the graph obtained in step 8, W f is the full-connection layer weight, softmax represents the full-connection layer classification operation,/>Representing the segmented result of the final output of the network.
Fig. 2 is a schematic diagram of a horizontal well section structure to which the present invention is directed. With the development of domestic shale gas and compact oil gas, the staged fracturing development mode of clustering perforation-composite bridge plug combination is widely applied in horizontal well construction. Firstly, a vertical shaft is drilled on the ground, the approximate position of a hydrocarbon reservoir is obtained through well logging detection, then a horizontal well is drilled, geological information such as gas measurement values, porosity, rock stratum distribution, predicted fracture sections and the like of the horizontal well is obtained through measurement, and then horizontal well sections which are located on the same small layer and have similar geological structures such as natural fracture development sections, similar reservoir parameters and similar rock mechanical parameters are divided into the same perforation fracturing sections according to the information, so that the horizontal well is divided into a plurality of fracturing sections. After the segmentation is finished, the construction can be carried out by using a cluster perforation-composite bridge plug combined operation process technology, wherein the first segment adopts an oil pipe, a continuous oil pipe and a cable crawler to carry out perforation and then fracturing, and the other segments adopt cables to put a combined instrument string into a well, so that the hydraulic pumping process technology is propelled in a mode of pumping with high inclination and horizontal well Duan Yongshui force. Firstly, a previous section is plugged by a composite bridge plug, then the section is subjected to shower hole splitting, a combined instrument string is started, and then the section is subjected to volume fracturing construction.
Fig. 3 is a schematic diagram of the neural network training process according to the present invention. Firstly, horizontal well data materials which are manually segmented before are acquired, the graph neural network takes the materials as a training set, and the geological data are subjected to graph modeling before training. And then creating weighted edges between the nodes of the graph according to geological information such as gas measurement values, porosity and the like and the distance relation between each segment, and finishing the creation of the input graph. And (3) performing the operation on all the horizontal well data in the training set to obtain an input diagram training set. Before training begins, the weights of the neural network and the weights of the full-connection layer are initialized, and then training can begin. Training of a graph is mainly divided into 5 steps: carrying out Laplace spectrum decomposition and normalization on the weighted adjacent matrix of the input graph to obtain a normalized Laplace matrix; solving an input graph node feature vector matrix; performing matrix multiplication operation on an input graph feature vector matrix, a normalized Laplace matrix and a graph neural network weight matrix, and performing nonlinear activation processing by using relu functions to update the hidden state of the graph nodes; repeating the previous process until all hidden layers of the graph neural network are subjected to iterative computation; inputting the final calculation result of the hidden layer into the full-connection layer for node classification, and dividing the result into one section, thereby obtaining a horizontal well segmentation scheme predicted by the graph neural network; calculating the difference between a network predicted segmentation scheme and a manual segmentation scheme by using a cross entropy loss function, and carrying out back propagation by using an Adam optimization algorithm to update the network weight to the direction of decreasing the loss function;
After one training is finished, whether the training is finished is judged, if not, the next graph is input to repeat the operation to perform the next training, and if all graphs in the training set are trained, the graph neural network training is finished.
Fig. 4 is a schematic diagram of a process for automatically segmenting a trained neural network according to the present invention. Firstly, obtaining logging geological data of a horizontal well with segments, then carrying out graph modeling on the horizontal well according to the method for manufacturing the training set, inputting the graph into a trained graph neural network for forward propagation calculation, and finally obtaining a network prediction segmentation scheme, thereby realizing automatic prediction of the graph neural network.
The horizontal well fracturing segment segmentation task aims at perforating fracturing segment segmentation of the horizontal well according to logging geological information, and belongs to a typical task relying on manual experience and manual characteristic extraction, so that the mapping relation between logging data and segmentation schemes can be learned through a neural network, and automatic segmentation of the horizontal well is realized. For data processed by the traditional neural network, the data must be Euclidean space data with a regular structure, the data nodes are ordered nodes with unchanged sizes, and the number of adjacent nodes of any one data node is the same. However, the feature is not provided for the segmented logging data for the horizontal well segments, the data does not have space translation property, and different data nodes are not mutually independent, and the logging data belongs to typical non-Euclidean space graph data, so the invention proposes to use a graph neural network specially processing the characteristics of the non-Euclidean space data for automatically segmenting the horizontal well fracturing segments. To achieve this objective, 4 basic processes are required:
carrying out proper mathematical abstraction on logging data and converting the logging data into a graph node matrix X;
processing input data X by using a graph neural network, and learning a mapping relation between a segmentation scheme and the input data;
Is the segmentation scheme of the network output, and f () is the mapping relation learned by the network.
Comparing the network segmentation scheme with the manual segmentation scheme, continuously learning the network, and continuously updating the network weight to ensure that the quality of the network segmentation scheme is continuously close to that of the manual segmentation scheme;
Y is a manual segmentation scheme that is used to segment the object, Is a segmentation scheme of network output,/>Is the difference between the two schemes, and the network f () is updated in the direction in which the difference is as small as possible.
Obtaining the trained graphic neural network. Because the training goal is to reduce the difference between the network output segmentation scheme and the artificial segmentation scheme, the finally obtained graph neural network can output the segmentation scheme with the quality similar to that of the artificial one
The horizontal well fracturing section segmentation method based on the graph neural network comprises two parts, wherein the first part is training of the graph neural network, and the second part is reasoning of the graph neural network, and is characterized in that:
the training of the graphic neural network is used for realizing the mapping of the geological information of the horizontal well to the segmentation scheme. The existing horizontal wells which are segmented manually are constructed as graphs, the graphs are used as training data sets, the graph neural network is trained through forward propagation and reverse propagation, and the mapping relation between the geological information of the horizontal well and the segmentation scheme is learned, so that the difference between the network segmentation scheme and the manual segmentation scheme is as small as possible, and finally the trained graph neural network is obtained.
And the reasoning of the graph neural network is used for automatically segmenting the horizontal well which is not manually segmented through the trained graph neural network without manual intervention. The horizontal well to be segmented is constructed as an input graph, the input graph is input into a trained graph neural network, and the output of the network is the segmentation scheme.
As shown in fig. 5, the horizontal well fracturing segment segmentation method based on the graph neural network has the following working procedures:
Step 1, initializing the weights of the graph neural network And full link layer weight W f; wherein/>The weights of the layer 1 ith neurons in the graph neural network are represented. The number of layers of the neural network should be set between 16-32 layers, and the effect may be poor if the number of layers of the network is too small, and the training time may be too long and the fitting may be caused. The weights of all neurons together form a weight matrix of the graph neural network. W f represents the full connection layer weight matrix, and f represents the full connection layer. Two weight matrices are initialized using a normal distribution random method such that each element value is between 0 and 1 and obeys a normal distribution.
Step 2, determining the minimum sectional length of the horizontal well and the total length of the horizontal well, enabling the minimum sectional length to be an initial value of a graph node, enabling each section to be an input graph node, and calculating the number of the graph nodes according to the following formula;
where total_length represents the total length of the horizontal well, min_length represents the minimum segment length, X num represents the number of graph nodes, and its value is equal to the total length of the horizontal well divided by the minimum segment length rounded up.
Step 3, creating weighted edges X (i, j) between the graph nodes according to geological information such as gas measurement values, porosities and the like of the segments represented by the graph nodes in the step 2 and the distance relation between each segment; x (i, j) represents the weight of the edge between the graph node i and the graph node j, the value of the weight is between-1 and 1, and the edge weight is larger if the distance between two nodes is closer and the geological information is closer, and vice versa. The collection of all graph nodes and all weighted edges X (i, j) form the input graph of the graph neural network.
And 4, creating a self-connection weighted adjacency matrix of the graph according to the input graph obtained in the step3. The number of rows and the number of columns of the matrix are the same, and are the number of nodes in the graph; elements of row i and column j of matrixThe values of (2) are shown in the following formula:
Wherein X (i, j) represents the weighted edge weight between the i-th node and the j-th node obtained in step 3.
Step 5, carrying out Laplace spectrum decomposition and normalization on the self-connected weighted adjacent matrix obtained in the step 4 by using the following formula to obtain a normalized Laplace matrix L;
In the middle of Self-connecting weighted adjacency matrix for the graph obtained in step 4,/>For/>Degree matrix of L is pair/>And carrying out Laplace spectrum decomposition and normalization to obtain the Laplace matrix.
And 6, obtaining a eigenvector matrix H 0 of the Laplace matrix L obtained in the step 5, wherein the solving process is as follows:
|λI-L|＝0
(λI-L)α＝0
H0＝[α1,α2,...,αn]
Wherein I is an identity matrix, lambda is a matrix eigenvalue to be solved, L is an input graph Laplacian matrix obtained in the step 4, and alpha is an eigenvector corresponding to the eigenvalue lambda. All possible values of α together form a feature vector matrix H 0. The matrix represents an initial hidden state of the node;
And 7, inputting the hidden state feature vector matrix H l and the normalized Laplace matrix L of the input diagram into a first layer in the diagram neural network for calculation to obtain a calculation result Z l, wherein l=0 in the first calculation, and the calculation is started from the first layer of the diagram neural network. The calculation method is as follows:
Zl＝LHlWl
wherein W l represents the weight of the neural network of the first layer graph.
Step 8, performing nonlinear activation processing on the result in the step 7 by using Relu activation functions, completing calculation of a single-layer graph neural network, updating hidden states from H l to H l+1, and updating l self-adding 1;
Hl+1＝Relu(Zl)
l＝l+1
Step 9, if l < the total layer number of the graph neural network updated in the step 8, jumping to the step 7, and executing the step 7 and the step 8 again, otherwise jumping to the step 10.
And step 10, inputting a final result obtained by the calculation of the graph neural network into a full-connection layer for node classification, wherein nodes which are classified into one type are used as one section, so that a horizontal well segmentation scheme predicted by the graph neural network is obtained. The full connection layer calculation formula is as follows:
In the middle of For the final calculation result of the neural network of the graph obtained in step 9, W f is the full-connection layer weight, softmax represents the full-connection layer classification operation,/>The segmentation result of the final network output is represented, and the nodes which are divided into the same class are one segment.
And step 11, calculating the difference between the network predicted segmentation scheme and the artificial segmentation scheme by using the cross entropy loss function. The cross entropy loss function is as follows:
Wherein Y represents a manual segmentation scheme, Representing the network segmentation scheme obtained in step 10, Y i and/>Representing each element in the two vectors separately.
After the cross entropy is obtained, the Adam optimization algorithm is used for back propagation, so that the network weight is updated towards the direction of decreasing the loss function. The Adam optimization algorithm is calculated as follows:
t←t+1
mt←β1·mt-1+(1-β1)·gt
Wherein t represents a time step t; f (θ) represents the objective function to be optimized, in the present invention the cross entropy loss function; θ represents a network weight parameter to be updated, θ t-1 represents a parameter of a previous time step, and θ t represents a parameter of a current time step updated after one iteration; beta 1 and beta 2 are two important super parameters of the algorithm, and the general values are 0.9 and 0.999 respectively; alpha is learning rate, generally the initial value is 0.01 or 0.001, and learning rate attenuation is carried out after 10-20 iterations; epsilon represents the optimization algorithm bias and generally takes on a value of 10 -8.
Step 12, if horizontal well data to be trained still exist, training is not finished yet, the step2 is skipped, and the step2 to the step 12 are repeated; otherwise, all training is completed, and the step is skipped to step 13;
step 13, training is finished, and a trained graphic neural network is obtained;
Step 14, measuring geological information of the horizontal well to be segmented;
Step 15, repeating the steps 2 to 11;
step 16, outputting a segmentation scheme of the horizontal well to be segmented by the graph neural network;
The above is only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited by this, and any modification made on the basis of the technical scheme according to the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (7)
1. The horizontal well fracturing section segmentation method is characterized by comprising the following steps of;
Step 1, initializing weights of a graph neural network and weights of a full-connection layer;
Step 2, constructing an input diagram according to the geological information of the horizontal well: determining the minimum sectional length of the horizontal well and the total length of the horizontal well, enabling the minimum sectional length to be the initial value of the graph nodes, enabling each section to be an input graph node, and calculating the number of the graph nodes; according to the horizontal well geological information of the segments represented by the graph nodes and the distance relation between each segment, creating weighted edges between the graph nodes to represent weights of the two graph nodes and edges between the graph nodes, wherein the collection of all graph nodes and all weighted edges forms an input graph of the graph neural network;
step 3, creating a self-connection weighted adjacent matrix of the input diagram, and carrying out Laplacian spectrum decomposition on the self-connection weighted adjacent matrix to obtain a Laplacian matrix;
Step 4, solving a feature vector matrix of the Laplace matrix, inputting the feature vector matrix into a graph neural network for forward propagation calculation, and obtaining a network prediction segmentation scheme: solving a feature vector matrix of the Laplace matrix, hiding the input diagram to the first state The layer eigenvector matrix and the normalized Laplace matrix are input to the/>, of the graph neural networkThe layer calculates to obtain a calculation result; performing nonlinear activation processing on the calculation result by using Relu activation functions to finish calculation of the single-layer graph neural network until the feature vector matrix of each layer of the graph neural network is updated; inputting all updated results of the graph neural network into a full-connection layer for node classification, wherein nodes classified into one type are used as one section, so that a horizontal well segmentation scheme predicted by the graph neural network is obtained;
step 5, acquiring a manual segmentation scheme of the horizontal well fracturing section predicted by the network, calculating a difference value between the segmentation scheme predicted by the network and the manual segmentation scheme by using a cross entropy loss function, and back-propagating an update weight by using an Adam optimization algorithm;
Step 6, circulating the steps 2 to 5 until all the geological information of the horizontal well is completely trained, and obtaining a final trained graphic neural network;
and 7, inputting geological information of the horizontal well to be segmented into the graphic neural network to obtain a horizontal well fracturing segment segmentation scheme.
2. The horizontal well fracturing segment segmentation method according to claim 1, wherein the calculation process of the number of graph nodes is:
In the middle of Representing the total length of the horizontal well,/>Representing the minimum segment length,/>Representing the number of graph nodes.
3. The method of claim 1, wherein in step 3, the self-connecting weighted adjacency matrix is the first matrixLine/>Column element/>The values of (2) are:
In the middle of Represents the/>, obtained in step 3Personal node and/>Edge weights with weights among the nodes;
The laplace matrix is:
In the middle of Self-connecting weighted adjacency matrix for the input graph obtained in step 3,/>For/>Degree matrix of L is pair/>And carrying out Laplace spectrum decomposition and normalization to obtain the Laplace matrix.
4. The horizontal well fracturing section segmentation method of claim 1, wherein in step 5, the cross entropy loss function is:
In the middle of Representing a manual segmentation scheme,/>Segmentation scheme representing network predictions,/>And/>Representing each element in the two vectors separately;
the Adam optimization algorithm comprises the following calculation processes:
Wherein t represents a time step t; Representing an objective function to be optimized, namely a cross entropy loss function; /(I) Representing the network weight parameters to be updated,/>Parameter representing last time step,/>A parameter representing a current time step updated after an iteration; /(I)And/>Is two important super parameters of the algorithm, and the value is/>, respectively；/>For learning rate, the initial value is/>Learning rate decay is carried out after 10-20 iterations; /(I)Represents the deviation of the optimization algorithm, and takes the value as/>。
5. A horizontal well fracturing segment segmentation system, comprising:
the weight initialization module is used for initializing the weights of the graphic neural network and the weights of the full-connection layer;
The input diagram construction module is used for constructing an input diagram according to the geological information of the horizontal well; determining the minimum sectional length of the horizontal well and the total length of the horizontal well, enabling the minimum sectional length to be the initial value of the graph nodes, enabling each section to be an input graph node, and calculating the number of the graph nodes; according to the horizontal well geological information of the segments represented by the graph nodes and the distance relation between each segment, creating weighted edges between the graph nodes to represent weights of the two graph nodes and edges between the graph nodes, wherein the collection of all graph nodes and all weighted edges forms an input graph of the graph neural network;
The Laplace matrix calculation module is used for creating a self-connection weighted adjacent matrix of the input graph, and carrying out Laplace spectrum decomposition on the self-connection weighted adjacent matrix to obtain a Laplace matrix;
The network prediction segmentation scheme calculation module is used for solving a feature vector matrix of the Laplace matrix, inputting the feature vector matrix into the graph neural network for forward propagation calculation, and obtaining a network prediction segmentation scheme: solving a feature vector matrix of the Laplace matrix, hiding the input diagram to the first state The layer eigenvector matrix and the normalized Laplace matrix are input to the/>, of the graph neural networkThe layer calculates to obtain a calculation result; performing nonlinear activation processing on the calculation result by using Relu activation functions to finish calculation of the single-layer graph neural network until the feature vector matrix of each layer of the graph neural network is updated; inputting all updated results of the graph neural network into a full-connection layer for node classification, wherein nodes classified into one type are used as one section, so that a horizontal well segmentation scheme predicted by the graph neural network is obtained;
the backward propagation updating weight module is used for acquiring an artificial segmentation scheme of the horizontal well fracturing section predicted by the network, calculating a difference value between the segmentation scheme predicted by the network and the artificial segmentation scheme by using a cross entropy loss function, and then using an Adam optimization algorithm to backward propagate updating weight;
The graph neural network training completion acquisition module is used for repeatedly inputting the calculation processes of the graph construction module, the Laplace matrix calculation module, the network prediction segmentation scheme calculation module and the back propagation updating weight module until all the horizontal well geological information is completely trained, and obtaining a graph neural network with the final training completed;
the horizontal well fracturing segment segmentation module is used for inputting the geological information of the horizontal well to be segmented into the graphic neural network to obtain a horizontal well fracturing segment segmentation scheme.
6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, carries out the steps of the horizontal well fracturing segment segmentation method according to any one of claims 1 to 4.
7. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the horizontal well fracturing segment segmentation method according to any of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110130584.4A CN112906760B (en) | 2021-01-29 | 2021-01-29 | Horizontal well fracturing segment segmentation method, system, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110130584.4A CN112906760B (en) | 2021-01-29 | 2021-01-29 | Horizontal well fracturing segment segmentation method, system, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112906760A CN112906760A (en) | 2021-06-04 |
| CN112906760B true CN112906760B (en) | 2024-05-03 |
Family
ID=76121685
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110130584.4A Active CN112906760B (en) | 2021-01-29 | 2021-01-29 | Horizontal well fracturing segment segmentation method, system, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112906760B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116291336B (en) * | 2023-02-14 | 2024-05-24 | 电子科技大学 | Automatic segmentation clustering system based on deep self-attention neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109799533A (en) * | 2018-12-28 | 2019-05-24 | 中国石油化工股份有限公司 | A kind of method for predicting reservoir based on bidirectional circulating neural network |
| CN109812247A (en) * | 2019-02-21 | 2019-05-28 | 中国石油天然气股份有限公司 | A kind of perforation and fracturing process improving fine and close grease horizontal well transformation degree |
| CN109993203A (en) * | 2019-02-23 | 2019-07-09 | 中国海洋石油集团有限公司 | A kind of well logging oil-gas-water layer intelligent interpretation evaluation method |
| CN110287516A (en) * | 2019-05-13 | 2019-09-27 | 西南石油大学 | A kind of shale gas reservoir pressure break horizontal well compressibility integrated evaluating method |
| CN110400006A (en) * | 2019-07-02 | 2019-11-01 | 中国石油化工股份有限公司 | Oil well output prediction technique based on deep learning algorithm |
| CN112085161A (en) * | 2020-08-20 | 2020-12-15 | 清华大学 | Graph neural network method based on random information transmission |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2810045A1 (en) * | 2012-03-21 | 2013-09-21 | Oiltool Engineering Services, Inc. | Multizone frac system |
| US9879514B2 (en) * | 2014-08-26 | 2018-01-30 | Gas Technology Institute | Hydraulic fracturing system and method |
-
2021
- 2021-01-29 CN CN202110130584.4A patent/CN112906760B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109799533A (en) * | 2018-12-28 | 2019-05-24 | 中国石油化工股份有限公司 | A kind of method for predicting reservoir based on bidirectional circulating neural network |
| CN109812247A (en) * | 2019-02-21 | 2019-05-28 | 中国石油天然气股份有限公司 | A kind of perforation and fracturing process improving fine and close grease horizontal well transformation degree |
| CN109993203A (en) * | 2019-02-23 | 2019-07-09 | 中国海洋石油集团有限公司 | A kind of well logging oil-gas-water layer intelligent interpretation evaluation method |
| CN110287516A (en) * | 2019-05-13 | 2019-09-27 | 西南石油大学 | A kind of shale gas reservoir pressure break horizontal well compressibility integrated evaluating method |
| CN110400006A (en) * | 2019-07-02 | 2019-11-01 | 中国石油化工股份有限公司 | Oil well output prediction technique based on deep learning algorithm |
| CN112085161A (en) * | 2020-08-20 | 2020-12-15 | 清华大学 | Graph neural network method based on random information transmission |
Non-Patent Citations (3)
| Title |
|---|
| 测井神经网络技术综述;刘全稳;石油地球物理勘探;31(S1);第64-70页 * |
| 页岩气水平井产量主控因素分析及产能预测;王忠东等;测井技术;第41卷(第5期);第577-582页 * |
| 页岩气水平井压裂细分段密分簇优化设计及应用;高东伟;;长江大学学报(自然科学版);16(04);第40-44页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112906760A (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334949B (en) | Image classifier construction method based on optimized deep convolutional neural network structure fast evolution | |
| CN110807544B (en) | Oil field residual oil saturation distribution prediction method based on machine learning | |
| Karkevandi-Talkhooncheh et al. | Application of hybrid adaptive neuro-fuzzy inference system in well placement optimization | |
| CN110619353B (en) | Multi-scale logging curve automatic identification method based on deep learning | |
| CN113610945B (en) | Ground stress curve prediction method based on hybrid neural network | |
| CN103324954A (en) | Image classification method based on tree structure and system using same | |
| Li et al. | Reservoir production prediction with optimized artificial neural network and time series approaches | |
| CN114723095A (en) | Missing well logging curve prediction method and device | |
| CN112906760B (en) | Horizontal well fracturing segment segmentation method, system, equipment and storage medium | |
| Raji et al. | A new approach in well placement optimization using metaheuristic algorithms | |
| CN116384554A (en) | Method and device for predicting mechanical drilling speed, electronic equipment and computer storage medium | |
| CN114117599A (en) | Shield attitude position deviation prediction method | |
| Wang et al. | A novel surrogate-assisted multi-objective optimization method for well control parameters based on tri-training | |
| CN108710734B (en) | Numerical simulation method and device based on grid adaptive encryption and coarsening technology | |
| CN117574755B (en) | Hierarchical multistage optimization method for fracturing parameters of horizontal well of shale reservoir well factory | |
| Zhuang et al. | Multi-objective optimization of reservoir development strategy with hybrid artificial intelligence method | |
| Huang et al. | A deep-learning-based graph neural network-long-short-term memory model for reservoir simulation and optimization with varying well controls | |
| CN113435128A (en) | Oil and gas reservoir yield prediction method and device based on condition generation type countermeasure network | |
| CN110619382A (en) | Convolution depth network construction method suitable for seismic exploration | |
| CN116432053A (en) | Multi-mode data representation method based on modal interaction deep hypergraph neural network | |
| CN115099406A (en) | Stratum pressure inversion method and device based on multivariate time sequence | |
| Khamehchi et al. | Prediction of maximum oil production by gas lift in an Iranian field using auto-designed neural network | |
| Pan et al. | Dynamic analysis of nonlinear civil engineering structures using artificial neural network with adaptive training | |
| Son et al. | Two-stage sampling scheme for a CNN-based well placement optimization of 3D benchmark reservoirs | |
| Li et al. | Digital construction of geophysical well logging curves using the LSTM deep-learning network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |