CN108509694A - A kind of proppant laying form prediction method based on BP neural network - Google Patents
A kind of proppant laying form prediction method based on BP neural network Download PDFInfo
- Publication number
- CN108509694A CN108509694A CN201810205015.XA CN201810205015A CN108509694A CN 108509694 A CN108509694 A CN 108509694A CN 201810205015 A CN201810205015 A CN 201810205015A CN 108509694 A CN108509694 A CN 108509694A
- Authority
- CN
- China
- Prior art keywords
- proppant
- neural network
- parameter
- input
- discharge capacity
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH DRILLING; MINING
- E21B—EARTH DRILLING, e.g. DEEP 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
- E21B43/267—Methods for stimulating production by forming crevices or fractures reinforcing fractures by propping
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
A kind of proppant laying form prediction method based on BP neural network, includes the following steps：Select sand than, proppant grain size, proppant density, operational discharge capacity lay the principal element of form to influence proppant, and select balance height, equilibration time, whaleback leading edge distance for the major parameter of description proppant laying form；Laying form of the proppant in crack under different parameters is simulated using results of fracture simulation experimental provision；It compiles experimental data and is divided into training sample and test samples；The parameter of BP neural network is set, is established from sand than, the Nonlinear Mapping relationship of proppant grain size, proppant density, operational discharge capacity to balance height, equilibration time, whaleback leading edge distance；Neural network is trained and is examined；Form is laid to new samples proppant to predict；The present invention realizes the prediction that form is laid to proppant using the powerful learning ability of BP neural network, easy to operate, is convenient for Oil Field application.
Description
Technical field
The present invention relates to petroleum works field, more particularly to it is pre- that a kind of proppant based on BP neural network lays form
Survey method is Oil Field proppant and construction parameter for laying form prediction of the proppant in fracturing process in crack
Selection provide foundation.
Technical background
Since conventional gas and oil resource can no longer meet social demand, exploitation of the people for unconventional reservoir in recent years
Dynamics is increasingly deepened.Hydraulic fracturing technology has become the exploitation very effective means of unconventional reservoir petroleum resources, however is
The sand packed fracture that high flow conductivity is formed in reservoir, must just be such that proppant is effectively filled in crack, therefore right
Migration rule of the proppant in crack and laying form, which carry out research, to be particularly important.Before this for proppant in crack
Laying form research be concentrated mainly on using results of fracture simulation experimental provision to proppant sedimentation migration process simulate；But
It is not only the influence of supported dose of type but also to be influenced by factors such as construction parameters in migration process due to proppant,
So meeting the actual proppant sedimentation migration rule mathematic(al) representation in scene it is difficult to set up, this is just greatly limited to support
The prediction of laying form of the agent in crack.
Invention content
In order to overcome the defect of the above-mentioned prior art, the purpose of the present invention is to provide a kind of branch based on BP neural network
It supports agent and lays form prediction method, form can be not only laid to proppant and predict but also trained nerve can be utilized
Network carries out weight analysis, to obtain the weight for influencing proppant and laying form principal element, further to establish proppant
Evaluation model provides the actual weight foundation of science, in turn avoids experiment input and complicated mathematical operation.
In order to overcome the defect of the above-mentioned prior art, the technical scheme is that：
A kind of proppant laying form prediction method based on BP neural network, includes the following steps：
Step 1：Select sand than, proppant grain size, proppant density, operational discharge capacity be to influence proppant to lay form
Major parameter selects balance height, equilibration time, whaleback leading edge distance to lay the major parameter of form for description proppant；
Step 2：Using a kind of parallel board slit simulation apparatus of fracturing fluid leak to proppant under different parameters in crack
In laying form simulated, compile experimental data, experimental result be divided into two groups, one group is used as training sample, one
Group is used as test samples；
Step 3：Selecting, which influences proppant, lays input layer of the major parameter of form as BP neural network, description branch
It is the output layer of BP neural network to support agent and lay the parameter of form, and neural network parameter is arranged, training sample is input to nerve
Network is trained, to establish input layer to output layer Nonlinear Mapping relationship；
Step 4：BP neural network training finish after, by test samples be input to trained neural network obtain it is pre-
It surveys as a result, MATLAB Program corrcoef is carried out pair between the numerical matrix and test samples truth value matrix of prediction result
Than judging, if similarity, model is effective if 0.9 or more, if similarity is unsatisfactory for this requirement and resets neural network parameter
With the number of hidden nodes until meeting required precision；
Step 5：By new samples sand than, proppant grain size, proppant density, the value of operational discharge capacity be input to neural network
Model lays form to proppant and predicts.
The specific method is as follows for the step three：
Using three layers of BP neural network, input layer includes sand ratio, proppant grain size, proppant density, operational discharge capacity four
Neuron node；Output layer includes three balance height, equilibration time, whaleback leading edge distance neuron nodes, the number of hidden nodes
Initial value is set as any integer within 3 to 12, to establish the Nonlinear Mapping relationship from input layer to output layer.
The neural network stated is not higher than using no less than 20 groups of experimental data with the learning rate not higher than 0.05
Centesimal error, it is adaptive Ir gradient descent methods to select training function, and iteration step length 5000 instructs network
Practice.
The beneficial effects of the invention are as follows：The present invention selects selection sand ratio, proppant grain size, proppant density, operational discharge capacity
To influence the major parameter that proppant lays form, balance height, equilibration time, whaleback leading edge distance are selected to describe proppant
The major parameter for laying form lays form to the proppant under different parameters using parallel board slit simulation experimental provision and carries out
Interpretation is divided into training sample and detection sample, establishes BP neural network model and network parameter is arranged by simulation, right
Network training and adjustment are to set up from sand ratio, proppant grain size, proppant density, operational discharge capacity to balance height, balance
Time, whaleback leading edge distance optimal Nonlinear Mapping relationship, this method is easy to operate, both can be used for analyze proppant fortune
Move the reliable basis that rule can also be used as Oil Field selection proppant and construction parameter.
Description of the drawings
Fig. 1 is the BP neural network structure chart of the present invention.
Specific implementation mode
Further make specific elaboration, but the present invention to technical scheme of the present invention in a manner of enumerating embodiment below
Be not limited to it is set forth below for embodiment.
Step 1：Select sand than, proppant grain size, proppant density, operational discharge capacity be to influence proppant to lay form
Major parameter selects balance height, equilibration time, whaleback leading edge distance to lay the major parameter of form for description proppant.
Before experiment starts, experiment parameter select viscosity for 1mPas slippery water fracturing fluids, sand ratio including 0.02,0.03,
0.04,0.05,0.06,0.08 6 kind, proppant grain size include 0.21mm, 0.32mm, 0.45mm, 0.64mm, four kinds, proppant
Density includes 1450kg/m3、1540kg/m3、1890kg/m3、2600kg/m3、2770kg/m3、2880kg/m3、3020kg/m3Seven
Kind, operational discharge capacity includes 4m3/h、5m3/h、6m3/h、7m3/h、8m3Five kinds of/h, is designed using the various combination of these parameters
Experiment.
Step 2：Using a kind of parallel board slit simulation apparatus of fracturing fluid leak to proppant under different parameters in crack
In laying form simulated, compile experimental data, experimental result be divided into two groups, one group is used as training sample, one
Group is used as test samples.
It is entitled using Patent No. 201420450355.6《A kind of parallel board slit simulation apparatus of fracturing fluid leak》
Described device and experiment pipeline connect, and check the leakproofness of pipeline valve and described device；Fracturing fluid is added in fluid reservoir,
Mixing pump is opened to start to stir；According to sand ratio, proppant grain size, proppant density determined by experimental designs, one is weighed
Quantitative proppant is slowly added to liquid storage and thinks in tank, and constantly stirs, and proppant is made to be mixed with fracturing fluid；Open data acquisition
System opens screw pump, and the discharge capacity of pump is gradually increased to discharge capacity is designed, ensures that load fluid flow speed stability is worked as and observes proppant heap
Product height reaches termination of pumping after balance height, and three record balance height, equilibration time, whaleback leading edge distance parameter values obtain table 1
Shown in experimental data, by first 20 groups of experimental data be used as training sample, last 5 groups of data are as test samples.
Step 3：Selecting, which influences proppant, lays input layer of the major parameter of form as BP neural network, description branch
It is the output layer of BP neural network to support agent and lay the parameter of form, and neural network parameter is arranged, training sample is input to nerve
Network is trained, to establish input layer to output layer Nonlinear Mapping relationship；
BP neural network is established, selects sand than, the input of proppant grain size, proppant density, discharge capacity as neural network
Layer neuron selects the output layer neuron of balance height, equilibration time, whaleback leading edge distance as neural network, output layer
Transmission function to hidden layer is set as S type tangent functions, and training function setup is adaptive lr gradient descent methods, to establish
For input layer to the Nonlinear Mapping relationship of output layer, Nonlinear Mapping relationship refers to Fig. 1,
Step 4：BP neural network training finish after, by test samples be input to trained neural network obtain it is pre-
It surveys as a result, the input layer data of training sample and output layer data is normalized in MATLAB Program, calls
MATLAB Neural Network Toolbox program bags establish network, and it is any integer within 3 to 12 that the number of hidden nodes initial value, which is arranged,
Value, learning rate 0.05, error 0.01, iteration step length 5000；Numerical value of the MATLAB Program corrcoef to prediction result
Comparison judgement is carried out between matrix and test samples truth value matrix, model is effective if 0.9 or more if similarity, if similarity is not
Meet this requirement and then resets neural network parameter and the number of hidden nodes until meeting required precision.
Using the training sample data of normalized, the neural network of foundation is trained, is changed until meeting error
In generation, terminates.
In order to examine and improve the precision of neural network, using MATLAB programs to the input layer data normalizing of test samples
Change and handle and be input to neural network, obtained result anti-normalization processing is obtained into predicted value, by prediction value matrix and inspection
Similarity comparison and the number of hidden nodes is adjusted between the truth value matrix of sample, finds similarity highest when the number of hidden nodes is 7, god
Best through network, concrete numerical value comparison is shown in Table 2.
Step 5：By new samples sand than, proppant grain size, proppant density, the value of operational discharge capacity be input to neural network
Model lays form to proppant and predicts；It can be to new samples proppant in different rows using established BP neural network
Laying form under amount is predicted that the process no longer needs to carry out simulated experiment, and the input of manpower and financial resources is greatly saved,
It is simple and effective to facilitate popularization.
1 proppant of table lays simulated experiment result data table
2 BP neural network true value of table and predicted value contrast table
The above is only the part preferred embodiment of the present invention, any to be familiar with the possible profit of researcher in this field
It is changed with above-mentioned technical proposal.Therefore, any simple modification or equally set that technical solution according to the present invention is carried out
It changes, belongs to the scope of protection of present invention.
Claims (3)
1. a kind of proppant based on BP neural network lays form prediction method, which is characterized in that include the following steps：
Step 1：Select sand than, proppant grain size, proppant density, operational discharge capacity for influence proppant lay form it is main
Parameter selects balance height, equilibration time, whaleback leading edge distance to lay the major parameter of form for description proppant；
Step 2：Using a kind of parallel board slit simulation apparatus of fracturing fluid leak to proppant under different parameters in crack
It lays form to be simulated, compiles experimental data, experimental result is divided into two groups, one group is used as training sample, one group of work
For test samples；
Step 3：Selecting, which influences proppant, lays input layer of the major parameter of form as BP neural network, describes proppant
The parameter for laying form is the output layer of BP neural network, and neural network parameter is arranged, training sample is input to neural network
Be trained, to establish input layer to output layer Nonlinear Mapping relationship；
Step 4：After BP neural network training finishes, test samples are input to trained neural network and obtain prediction knot
Fruit, MATLAB Program corrcoef carry out comparison between the numerical matrix and test samples truth value matrix of prediction result and sentence
Disconnected, if similarity, model is effective if 0.9 or more, if similarity is unsatisfactory for this requirement and resets neural network parameter and hidden
Node layer number is until meeting required precision；
Step 5：By new samples sand than, proppant grain size, proppant density, the value of operational discharge capacity be input to neural network model
Form is laid to proppant to predict.
2. a kind of proppant based on BP neural network according to claim 1 lays form prediction method, feature exists
In the specific method is as follows for the step three：
Using three layers of BP neural network, input layer includes four sand ratio, proppant grain size, proppant density, operational discharge capacity nerves
First node；Output layer includes three balance height, equilibration time, whaleback leading edge distance neuron nodes, the number of hidden nodes initial value
Any integer being set as within 3 to 12, to establish the Nonlinear Mapping relationship from input layer to output layer.
3. a kind of proppant based on BP neural network according to claim 1 or 2 lays form prediction method, feature
It is, the neural network is not higher than percent using no less than 20 groups of experimental data with the learning rate not higher than 0.05
One error, it is adaptive Ir gradient descent methods to select training function, and iteration step length 5000 is trained network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810205015.XA CN108509694B (en) | 2018-03-13 | 2018-03-13 | Proppant paving form prediction method based on BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810205015.XA CN108509694B (en) | 2018-03-13 | 2018-03-13 | Proppant paving form prediction method based on BP neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108509694A true CN108509694A (en) | 2018-09-07 |
| CN108509694B CN108509694B (en) | 2022-02-18 |
Family
ID=63376548
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810205015.XA Active CN108509694B (en) | 2018-03-13 | 2018-03-13 | Proppant paving form prediction method based on BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108509694B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063403A (en) * | 2018-10-22 | 2018-12-21 | 西安石油大学 | A kind of slippery water Optimized fracturing design method |
| CN110965977A (en) * | 2019-11-20 | 2020-04-07 | 中国石油大学(北京) | Fracturing construction analysis method |
| CN111622730A (en) * | 2020-05-29 | 2020-09-04 | 中国石油大学(华东) | Fracturing sand adding design method based on large-scale parallel plate proppant migration and placement model experiment |
| CN112761609A (en) * | 2021-02-19 | 2021-05-07 | 西南石油大学 | Optimization method for efficient laying of propping agent in hydraulic fracturing operation |
| CN113095398A (en) * | 2021-04-08 | 2021-07-09 | 西南石油大学 | Fracturing data cleaning method of BP neural network based on genetic algorithm optimization |
| CN115199238A (en) * | 2022-09-15 | 2022-10-18 | 四川省贝特石油技术有限公司 | Method and system for controlling feeding of superfine temporary plugging agent for gas reservoir exploitation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008080093A2 (en) * | 2006-12-21 | 2008-07-03 | Verenium Corporation | Amylases and glucoamylases, nucleic acids encoding them and methods for making and using them |
| US20150039544A1 (en) * | 2013-07-31 | 2015-02-05 | Schlumberger Technology Corporation | Resource production forecasting |
| CN106703773A (en) * | 2015-08-03 | 2017-05-24 | 中国矿业大学 | Liquid carbon dioxide bomb, propping agent blasting anti-reflection method and device |
| CN107194068A (en) * | 2017-05-22 | 2017-09-22 | 中国石油大学(北京) | Shale gas fracturing process underground unusual service condition real-time estimate method for early warning and device |
-
2018
- 2018-03-13 CN CN201810205015.XA patent/CN108509694B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008080093A2 (en) * | 2006-12-21 | 2008-07-03 | Verenium Corporation | Amylases and glucoamylases, nucleic acids encoding them and methods for making and using them |
| US20150039544A1 (en) * | 2013-07-31 | 2015-02-05 | Schlumberger Technology Corporation | Resource production forecasting |
| CN106703773A (en) * | 2015-08-03 | 2017-05-24 | 中国矿业大学 | Liquid carbon dioxide bomb, propping agent blasting anti-reflection method and device |
| CN107194068A (en) * | 2017-05-22 | 2017-09-22 | 中国石油大学(北京) | Shale gas fracturing process underground unusual service condition real-time estimate method for early warning and device |
Non-Patent Citations (4)
| Title |
|---|
| MING LI等: "Research for Unsteady Seepage Flow of Asymmetrical Fractured Vertical Wells in Coalbed", 《2017 2ND INTERNATIONAL CONFERENCE ON NEW ENERGY AND RENEWABLE RESOURCES 》 * |
| SU YUJIE等: "The applications of BP neural network based on MIV in hydraulic fracturing", 《ADVANCED MATERIALS RESEARCH》 * |
| 周德胜等: "滑溜水压裂主裂缝内支撑剂输送规律实验及数值模拟", 《石油钻采工艺》 * |
| 唐伏平等: "考虑壁面影响的矩形裂缝通道中的颗粒沉降", 《排灌机械工程学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063403A (en) * | 2018-10-22 | 2018-12-21 | 西安石油大学 | A kind of slippery water Optimized fracturing design method |
| CN109063403B (en) * | 2018-10-22 | 2022-09-13 | 西安石油大学 | Optimal design method for slickwater fracturing |
| CN110965977A (en) * | 2019-11-20 | 2020-04-07 | 中国石油大学(北京) | Fracturing construction analysis method |
| CN111622730A (en) * | 2020-05-29 | 2020-09-04 | 中国石油大学(华东) | Fracturing sand adding design method based on large-scale parallel plate proppant migration and placement model experiment |
| CN111622730B (en) * | 2020-05-29 | 2022-04-01 | 中国石油大学(华东) | Fracturing sand adding design method based on large-scale parallel plate proppant migration and placement model experiment |
| CN112761609A (en) * | 2021-02-19 | 2021-05-07 | 西南石油大学 | Optimization method for efficient laying of propping agent in hydraulic fracturing operation |
| CN113095398A (en) * | 2021-04-08 | 2021-07-09 | 西南石油大学 | Fracturing data cleaning method of BP neural network based on genetic algorithm optimization |
| CN113095398B (en) * | 2021-04-08 | 2022-07-12 | 西南石油大学 | Fracturing data cleaning method of BP neural network based on genetic algorithm optimization |
| CN115199238A (en) * | 2022-09-15 | 2022-10-18 | 四川省贝特石油技术有限公司 | Method and system for controlling feeding of superfine temporary plugging agent for gas reservoir exploitation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108509694B (en) | 2022-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108509694A (en) | A kind of proppant laying form prediction method based on BP neural network | |
| RU2705012C1 (en) | Systems and methods for estimating and optimizing efficiency of stimulation using bypass devices | |
| Hamidpour et al. | Experimental study of some important factors on nonwetting phase recovery by cocurrent spontaneous imbibition | |
| CN108547612B (en) | Method and device for determining oil yield and gas yield of shale oil in-situ conversion | |
| CN107808032B (en) | Shale gas reservoir water saturation calculation method and system | |
| Cao | Development of a two-phase flow coupled capacitance resistance model | |
| CN107301483A (en) | The rapid integrated method for evaluating non-producing reserves economic producing feasibility | |
| CN108798657A (en) | Logging explanation method based on drilling fluid logging parameter Gas Logging Value | |
| Lee et al. | Investigation of consolidation-induced solute transport. II: Experimental and numerical results | |
| CN106875286A (en) | A kind of polymer flooding oil field overall process notes poly- parameter hierarchy optimization decision-making technique | |
| Shen et al. | Spontaneous imbibition in asymmetric branch-like throat structures in unconventional reservoirs | |
| Fattah et al. | Improved oil formation volume factor (Bo) correlation for volatile oil reservoirs: An integrated non-linear regression and genetic programming approach | |
| CN107133879B (en) | Method for screening similar oil fields | |
| Fadaei et al. | A modified semi-empirical correlation for designing two-phase separators | |
| CN111494988B (en) | Deep learning-based crude oil storage tank water cutting method and equipment | |
| Fattakhov et al. | Using artificial neural networks for analyzing efficiency of advanced recovery methods | |
| Rostami et al. | Proposing a rigorous empirical model for estimating the bubble point pressure in heterogeneous carbonate reservoirs | |
| CN107808374B (en) | Numerical method for evaluating fracturing effect of oil and gas reservoir | |
| CN106802986A (en) | CO being taken turns a kind of viscous crude bottom and edge water more2Handle up evaluation method | |
| CN106920037B (en) | Heavy oil edge bottom water reservoir multi-wheel CO based on water content2Huff and puff well selection method | |
| Gan et al. | Wax-Out Cryo-trapping: A New Trap-Filling Process in Fluid Migration to Oilfields | |
| CN107358374A (en) | A kind of pressure break decision-making technique based on test data and flow unit | |
| CN115705452A (en) | Novel recovery ratio prediction method for middle and later stages of integrated sandstone reservoir development | |
| Chai et al. | Integrated Hierarchy–Correlation Model for Evaluating Water-Driven Oil Reservoirs | |
| CN105781536A (en) | Method for measuring and calculating petroleum reservoir porosity degree |
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 |