CN109002927A - Oil field prospecting reserve forecasting method based on recurrent neural network - Google Patents
Oil field prospecting reserve forecasting method based on recurrent neural network Download PDFInfo
- Publication number
- CN109002927A CN109002927A CN201810855341.5A CN201810855341A CN109002927A CN 109002927 A CN109002927 A CN 109002927A CN 201810855341 A CN201810855341 A CN 201810855341A CN 109002927 A CN109002927 A CN 109002927A
- Authority
- CN
- China
- Prior art keywords
- data
- recurrent neural
- neural network
- oil field
- reserves
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 91
- 230000000306 recurrent effect Effects 0.000 title claims abstract description 83
- 238000013277 forecasting method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 53
- 238000000034 method Methods 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 9
- 230000000644 propagated effect Effects 0.000 claims description 9
- 238000011161 development Methods 0.000 claims description 6
- 230000001154 acute effect Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 3
- 241001269238 Data Species 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 239000000463 material Substances 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 244000007853 Sarothamnus scoparius Species 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Mining & Mineral Resources (AREA)
- Animal Husbandry (AREA)
- Agronomy & Crop Science (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Marine Sciences & Fisheries (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of oil field prospecting reserve forecasting method based on recurrent neural network, and the oil field prospecting reserve forecasting method based on recurrent neural network of being somebody's turn to do includes: step 1, obtains the basic data of recurrent neural network；Step 2, basic data is pre-processed, obtains the basic data of preliminary treatment；Step 3, using the inner link between basic data, the integrated data that recurrent neural network modeling needs are generated；Step 4, the training data by above-mentioned data as recurrent neural network obtains the recurrent neural networks model for prediction；Step 5, the input prediction annualized basis data in recurrent neural networks model generate prediction output.The drawbacks of artificial design predictive equation of the oil field prospecting reserve forecasting method abandoning tradition based on recurrent neural network, learn the prediction model for being best suited for the reserves data out from a large amount of historical data using computer, realizes the accurate prediction for oil field prospecting reserves.
Description
Technical field
The present invention relates to oil field development technical fields, especially relate to a kind of oil field prospecting based on recurrent neural network
Reserve forecasting method.
Background technique
Deep learning is an important algorithm in machine learning about data modeling, wherein RNN prediction model is depth
Learn the classic applications in time series field modeling, RNN possesses to complexity the multi-level mapping of training data with its network
The modeling ability of data has a wide range of applications in data modeling field.
Why RNN (recurrent neural network) is known as circulation neural network, i.e. a sequence current output and front
It exports also related.The specific form of expression is that network can remember the information of front and be applied to the calculating currently exported
In, i.e., the node between hidden layer is no longer connectionless but has connection, and the input of hidden layer not only includes input layer
Output further includes the output of last moment hidden layer.Theoretically, RNN can be handled the sequence data of any length.But
It is in practice, often to assume that current state is only related to several states of front to reduce complexity.
Oil field prospecting reserve forecasting causes to predict that difficulty is big due to its market factor, development cost, the factors such as artificial decision
Big to increase, traditional reserve forecasting model such as gompertz model, Master Weng's cyclic model, broom shape model etc. can only fitting data variation
It is substantially regular, error of fitting is larger, and S sigmoid growth curve activation primitive is as non-as known to each node of RNN model
Linear Mapping unit can theoretically be fitted arbitrarily complicated equation, therefore have outstanding modeling ability.We invent thus
A kind of new oil field prospecting reserve forecasting method based on recurrent neural network, solves the above technical problem.
Summary of the invention
Learn to be best suited for the storage out from a large amount of historical data using computer the object of the present invention is to provide a kind of
The prediction model of data is measured, realizes and the oil field prospecting based on recurrent neural network of oil field prospecting reserves accurately predicted is stored up
Measure prediction technique.
The purpose of the present invention can be achieved by the following technical measures: the oil field prospecting reserves based on recurrent neural network are pre-
Survey method, the oil field prospecting reserve forecasting method based on recurrent neural network of being somebody's turn to do includes: step 1, obtains recurrent neural network
Basic data；Step 2, basic data is pre-processed, obtains the basic data of preliminary treatment；Step 3, basic data is utilized
Between inner link, generate recurrent neural network modeling need integrated data；Step 4, it is used as and is passed by above-mentioned data
The training data for returning neural network obtains the recurrent neural networks model for prediction；Step 5, in recurrent neural networks model
Middle input prediction annualized basis data generate prediction output.
The purpose of the present invention can be also achieved by the following technical measures:
In step 1, the basic data based on the existing each block in oil field over the years, obtains each block in oil field over the years
Resource discovery degree, accumulative oil reservoir number, adds up proved reserves data in year at prospect pit degree of prospecting, and above-mentioned data are made recurrence mind
Through network foundation data.
In step 2, original basic data include format not to, sequence not to, duplicate data, to original basis
Data screening, filtering obtain the basic data of preliminary treatment.
In step 3, it based on the basic data of preliminary treatment, using the inner link between basic data, takes
The mode be added, subtract each other, be multiplied, being divided by generates the integrated data that recurrent neural network modeling needs.
In step 4, it is predicted in recurrent neural networks model using time series data, passes through establishment and analysis time sequence
Column, development process, direction and the trend reflected according to time series are analogized or are extended, so as to next section of prediction
The level being likely to be breached in time or later several years, comprising: collect and arrange the historical data data of certain changing rule；It is right
These data informations carry out inspection identification, line up ordered series of numbers；Analysis time ordered series of numbers, therefrom find the changing rule change over time and
The rule of variation obtains certain mode.
In step 4, to reflect that the single geologic element in oil field explores the inherent changing rule between reserves data over the years, benefit
It is training data with the history of the single geologic element in oil field exploration reserves data, training data is carried out using recurrent neural network
Recurrence learning is predicted according to exploration reserves of the model parameter learnt to the single geologic element in prediction time.
In step 4, the oil field prospecting reserve forecasting method as a kind of based on recurrent neural network, the single geology in oil field
Unit explores the reflection of the inherent changing rule between reserves data over the years:
The single geologic element in oil field explores reserves data over the years has the characteristics that typical time series data, i.e. oil field
Annual exploration reserves data there is periodically variable characteristic, and there is no acute variation, in addition, exploration reserves data
It is also influenced by the investment of current year and workload, it is over the years to there is the single geologic element of random perturbation namely oil field in the reasonable scope
It is the presence of inherent changing rule between exploration reserves data.
In step 4, recurrent neural network study training data description used are as follows:
X={ x1,x2,...,xn}
xnFor the column vector of 3 dimensions, as a training sample, X is the set of training sample；
The loss function of recurrence learning is carried out to training data for recurrent neural network, wherein t is the time, and j is training sample
This serial number, yt,jFor the reserves true value of t time j training sample,For the model predication value of t time j training sample, V is instruction
Practice the quantity of sample, J indicates entropy, and θ is the parameter that can learn, the learning objective as network；
It is the propagated forward process of network for the traffic propagation of recurrent neural network forward recursion procedure.Wherein, h table
Show the number of plies of current recursion neural network, h ' indicates that the layer before h layers, i indicate the serial number of training sample, and k indicates next layer
The number of plies, H indicates the number of plies before h layer, and I indicates the number of training sample, a h layers of output valve of expression, θhH layers of expression swashs
Function living, b indicate the output valve of activation primitive, and ω indicates the parameter that network layer can learn；
For the output of output layer, variable meaning is identical as preceding formula；
For the back-propagation process of error；Wherein t indicates the time, and h indicates the current number of plies,H layers of expression is in t moment
Gradient, θ ' indicate the derivative of activation primitive.
In step 4, each input sample of recurrent neural network is prospect pit degree of prospecting, and resource discovery degree tires out
Oil reservoir number is counted, training label is the accumulative proved reserves data of current year, to model to the sequence data, by every 3 years
Training sample of the training data as a batch, obtains the changing rule between the data in continuous time；In entire model,
The information flow of one one-way flow is to reach hidden unit, the information of another at the same time one-way flow from input unit
Stream reaches output unit from hidden unit；Meanwhile the parameter learning of recurrent neural network is as traditional neural network algorithm,
It is equally using back-propagation algorithm, but if recurrent neural network is carried out network expansion, then input layer is to hidden layer
The data of parameter, hidden layer to output layer between parameter, hidden layer are shared, and in using gradient descent algorithm,
The output of each step not only relies on the network currently walked, and also relies on the state of several step networks in front.
In steps of 5, in forecast year, enterprise formulates production plan, the budget of workload, investment comprising investment,
Using the budget of the workload of investment, investment as the input data of recurrent neural networks model, by recurrent neural network root
According to the workload of basic data and investment, the value of investment, pass through the propagated forward process of recurrent neural network network, prediction
Generate the numerical value of annual prediction proved reserves.
The oil field prospecting reserve forecasting method based on recurrent neural network in the present invention, is related to planning in oil field prospecting
It is the oil field prospecting planning stage using the oil and gas reserves forecasting problem of RNN (recurrent neural network) model of deep learning in journey
Exploration plan provide decision references.This method is to reflect that the single geologic element in oil field explores the inherence between reserves data over the years
Changing rule is training data using the history exploration reserves data of the single geologic element in oil field, using recurrent neural network pair
Training data carry out recurrence learning, according to model parameter learn to predict the time single geologic element exploration reserves into
Row prediction.The single geologic element in oil field explores reserves data over the years has the characteristics that typical time series data, i.e. oil field
Annual exploration reserves data there is periodically variable characteristic, and there is no acute variation, in addition, exploration reserves data
It is also influenced by the investment of current year and workload, it is over the years to there is the single geologic element of random perturbation namely oil field in the reasonable scope
It is the presence of inherent changing rule between exploration reserves data.The drawbacks of artificial design predictive equation of the method abandoning tradition,
Learn the prediction model for being best suited for the reserves data out from a large amount of historical data using computer, is based on passing by one kind
Return the oil field prospecting reserve forecasting method of neural network, realizes the accurate prediction for oil field prospecting reserves.
Detailed description of the invention
Fig. 1 is the stream of a specific embodiment of the oil field prospecting reserve forecasting method of the invention based on recurrent neural network
Cheng Tu；
Fig. 2 is RNN network propagated forward schematic diagram in a specific embodiment of the invention.
Specific embodiment
To enable above and other objects, features and advantages of the invention to be clearer and more comprehensible, preferably implementation is cited below particularly out
Example, and cooperate shown in attached drawing, it is described in detail below.
The process for the oil field prospecting reserve forecasting method based on recurrent neural network that as shown in FIG. 1, FIG. 1 is of the invention
Figure.
Each block in oil field over the years is obtained based on the basic data of the existing each block in oil field over the years in step 101
Resource discovery degree, accumulative oil reservoir number, adds up proved reserves data in year at prospect pit degree of prospecting, using above-mentioned data as RNN
The basic data of (recurrent neural network).
In step 102, in the basic data processing of each block in oil field over the years, original basic data include format not
, sequence to, duplicate data, cannot not used directly in prediction task, need to screen original basic data, mistake
Filter, obtains the basic data of preliminary treatment, solves the problems, such as Data duplication, mixed and disorderly.
In step 103, the integrated data that RNN modeling needs are not included in the basic data of preliminary treatment.Tentatively to locate
Based on the basic data of reason, using the inner link between basic data, the mode for taking addition, subtracting each other, be multiplied, being divided by,
Generate RNN modeling need integrated data, solve the problems, such as present in basic data repeat, it is mixed and disorderly, solve basis
The data problem incompatible with RNN model.
In step 104, training data by above-mentioned data as RNN obtains the final RNN model for prediction.
Time series data forecasting problem can be briefly described for by establishment and analysis time sequence, according to time series
Development process, direction and the trend reflected, is analogized or is extended, so as to predicting lower a period of time or later several years
The level being inside likely to be breached.Its content includes: collection and the historical data data for arranging certain changing rule；These data are provided
Material carries out inspection identification, lines up ordered series of numbers；Analysis time ordered series of numbers therefrom finds the rule that the changing rule changes over time and changes
Rule, obtains certain mode.
The purpose of RNN is for processing sequence data, and why RNN is known as recurrent neural network, i.e. a sequence is current
Output and front output it is also related.The specific form of expression be network can the information to front remember and be applied to work as
In the calculating of preceding output, i.e., the node between hidden layer is no longer connectionless but has connection, and the input of hidden layer is not only
Output including input layer further includes the output of last moment hidden layer.Theoretically, RNN can be to the sequence data of any length
It is handled.
Understandable in order to become apparent the propagated forward mode of RNN, we enumerate in a specific embodiment of the invention
RNN network propagated forward schematic diagram, as shown in Figure 2:
The inherent changing rule between reserves data is explored over the years for the reflection single geologic element in oil field, it is single using oil field
The history exploration reserves data of geologic element are training data, carry out recurrence learning to training data using recurrent neural network,
It is predicted according to exploration reserves of the model parameter learnt to the single geologic element in prediction time.
As a kind of oil field prospecting reserve forecasting method based on recurrent neural network, the single geologic element in oil field is surveyed over the years
Visit the inherent changing rule reflection between reserves data:
The single geologic element in oil field explores reserves data over the years has the characteristics that typical time series data, i.e. oil field
Annual exploration reserves data there is periodically variable characteristic, and there is no acute variation, in addition, exploration reserves data
It is also influenced by the investment of current year and workload, it is over the years to there is the single geologic element of random perturbation namely oil field in the reasonable scope
It is the presence of inherent changing rule between exploration reserves data.
As a kind of oil field prospecting reserve forecasting method based on recurrent neural network, used in recurrent neural network study
Training data can be described as:
X={ x1,x2,...,xn}
xnFor the column vector of 3 dimensions, as a training sample, X is the set of training sample；
The loss function of recurrence learning is carried out to training data for recurrent neural network, wherein t is the time, and j is training sample
This serial number, yt,jFor the reserves true value of t time j training sample,For the model predication value of t time j training sample, V is instruction
Practice the quantity of sample, J indicates entropy, and θ is the parameter that can learn, the learning objective as network；
It is the propagated forward process of network for the traffic propagation of recurrent neural network forward recursion procedure.Wherein, h table
Show the number of plies of current RNN, h ' indicates that the layer before h layers, i indicate the serial number of training sample, and H indicates the number of plies before h layers, I
Indicate the number of training sample, a indicates h layers of output valve, θhIndicate that h layers of activation primitive, b indicate the output of activation primitive
Value, ω indicate the parameter that network layer can learn.
For the output of output layer, variable meaning is identical as preceding formula；
For the back-propagation process of error；Wherein t indicates the time, and h indicates the current number of plies,H layers of expression is in t moment
Gradient, θ ' indicate the derivative of activation primitive, and other parameters are identical as preceding formula.
Optimal recurrent neural network is obtained by exploring the training dataset training that reserves data form over the years by oil field,
Then the investment and workload that current year is combined according to the input data in prediction time, input trained recurrent neural network mould
Type.So far, recurrent neural network can go out to predict that reserves are explored in the prediction of the oil field geologic unit in time with automatic Prediction.
The time series models gone out by RNN model modeling, which have, is able to reflect historical data real change rule, excludes
The features such as human factor is good to the subjective impact and reusability of model, plays in modern time series modeling field and focuses on
The effect wanted.Nowadays, oil field prospecting reserves data are as typical time series data, but its data variation rule is by oil gas
The influence of the factors such as the market price, managerial decision, oil-gas exploration and development degree causes to predict such data using traditional equation
Difficulty greatly increases, and the exploration decision led is restricted by the result of decision before, causes a vicious circle.The present invention
Be related to a kind of oil field prospecting reserve forecasting method based on recurrent neural network, more particularly to data variation rule by it is a variety of not
In the case of being influenced with factor, oil field prospecting reserves are predicted using recurrent neural network, ensure that model modeling result
To the objectivity that data changing rule extracts, and the interference of Outlier Data point can be resisted to a certain extent, realize that oil field is surveyed
Visit the accuracy of reserve forecasting data.
In the oil field prospecting reserve forecasting method based on recurrent neural network, each input sample of RNN network is
(prospect pit degree of prospecting, resource discovery degree add up oil reservoir number), training label is the accumulative proved reserves data of current year, is
The sequence data is modeled, using every 3 years training datas as the training sample of a batch, obtains the continuous time
Changing rule between data.In entire model, the information flow of an one-way flow is to reach to hide list from input unit
Member, the information flow of another at the same time one-way flow reaches output unit from hidden unit.Meanwhile the parameter learning of RNN
It is equally using BP (backpropagation) algorithm, but with some difference as traditional neural network algorithm.If by RNN
Network expansion is carried out, then the data of input layer to parameter, hidden layer to output layer between the parameter of hidden layer, hidden layer are
Shared, and no for traditional neural network.And using in gradient descent algorithm, the output of each step, which not only relies on, works as
The network of preceding step, and also rely on the state of several step networks in front.
In step 105, input prediction annualized basis data generate prediction output.The RNN model that abovementioned steps training is completed
Include the parameter succeeded in school.In forecast year, enterprise formulates production plan, the budget of workload, investment comprising investment.It will
Input data of the budget of the workload, investment of investment as RNN model, by RNN model according to basic data and throwing
The value of the workload, investment that enter, by the propagated forward process of RNN network, prediction generates the number of annual prediction proved reserves
Value, provides reference for business decision.
Claims (10)
1. the oil field prospecting reserve forecasting method based on recurrent neural network, which is characterized in that should be based on recurrent neural network
Oil field prospecting reserve forecasting method includes:
Step 1, the basic data of recurrent neural network is obtained；
Step 2, basic data is pre-processed, obtains the basic data of preliminary treatment；
Step 3, using the inner link between basic data, the integrated data that recurrent neural network modeling needs are generated；
Step 4, the training data by above-mentioned data as recurrent neural network obtains the recurrent neural network mould for prediction
Type；
Step 5, the input prediction annualized basis data in recurrent neural networks model generate prediction output.
2. the oil field prospecting reserve forecasting method according to claim 1 based on recurrent neural network, which is characterized in that
In step 1, based on the basic data of the existing each block in oil field over the years, the resource for obtaining each block in oil field over the years verifies journey
Degree, accumulative oil reservoir number, adds up proved reserves data in year at prospect pit degree of prospecting, and above-mentioned data are made recurrent neural network basis
Data.
3. the oil field prospecting reserve forecasting method according to claim 1 based on recurrent neural network, which is characterized in that
In step 2, original basic data include format not to, sequence not to, duplicate data, to the screening of original basic data,
Filtering, obtains the basic data of preliminary treatment.
4. the oil field prospecting reserve forecasting method according to claim 1 based on recurrent neural network, which is characterized in that
In step 3, based on the basic data of preliminary treatment, using the inner link between basic data, take addition, subtract each other,
The mode be multiplied, being divided by generates the integrated data that recurrent neural network modeling needs.
5. the oil field prospecting reserve forecasting method according to claim 1 based on recurrent neural network, which is characterized in that
In step 4, in recurrent neural networks model using time series data predict, by establishment and analysis time sequence, according to when
Between the sequence development process, direction and the trend that are reflected, analogized or extended, so as to predict lower a period of time or after
The level being likely to be breached in several years, comprising: collect and arrange the historical data data of certain changing rule；These data are provided
Material carries out inspection identification, lines up ordered series of numbers；Analysis time ordered series of numbers therefrom finds the rule that the changing rule changes over time and changes
Rule, obtains certain mode.
6. the oil field prospecting reserve forecasting method according to claim 5 based on recurrent neural network, which is characterized in that
It is single using oil field to reflect that the single geologic element in oil field explores the inherent changing rule between reserves data over the years in step 4
The history exploration reserves data of geologic element are training data, carry out recurrence learning to training data using recurrent neural network,
It is predicted according to exploration reserves of the model parameter learnt to the single geologic element in prediction time.
7. the oil field prospecting reserve forecasting method according to claim 6 based on recurrent neural network, which is characterized in that
In step 4, as a kind of oil field prospecting reserve forecasting method based on recurrent neural network, the single geologic element in oil field is surveyed over the years
Visit the inherent changing rule reflection between reserves data:
The single geologic element in oil field explores reserves data over the years has the characteristics that typical time series data, i.e., oil field is every
Year exploration reserves data have periodically variable characteristic, and there is no acute variation, in addition, exploration reserves data also by
The investment of current year and workload influence, and there is the single geologic element of random perturbation namely oil field in the reasonable scope and explore over the years
It is the presence of inherent changing rule between reserves data.
8. the oil field prospecting reserve forecasting method according to claim 7 based on recurrent neural network, which is characterized in that
In step 4, recurrent neural network study training data description used are as follows:
X={ x1,x2,...,xn}
xnFor the column vector of 3 dimensions, as a training sample, X is the set of training sample；
The loss function of recurrence learning is carried out to training data for recurrent neural network, wherein t is the time, and j is training sample sequence
Number, yt,jFor the reserves true value of t time j training sample,For the model predication value of t time j training sample, V is training sample
This quantity, J indicate entropy, and θ is the parameter that can learn, the learning objective as network；
It is the propagated forward process of network for the traffic propagation of recurrent neural network forward recursion procedure.Wherein, h expression is worked as
The number of plies of preceding recurrent neural network, h ' indicate that the layer before h layers, i indicate the serial number of training sample, and k indicates next layer of layer
Number, H indicate the number of plies before h layers, and I indicates the number of training sample, and a indicates h layers of output valve, θhIndicate h layers of activation letter
Number, b indicate the output valve of activation primitive, and ω indicates the parameter that network layer can learn；
For the output of output layer, variable meaning is identical as preceding formula；
For the back-propagation process of error；Wherein t indicates the time, and h indicates the current number of plies,Indicate the h layers of gradient in t moment,
The derivative of θ ' expression activation primitive.
9. the oil field prospecting reserve forecasting method according to claim 8 based on recurrent neural network, which is characterized in that
In step 4, each input sample of recurrent neural network is prospect pit degree of prospecting, and resource discovery degree adds up oil reservoir number,
Training label is the accumulative proved reserves data of current year, and to model to the sequence data, every 3 years training datas are made
For the training sample of a batch, the changing rule between the data in continuous time is obtained；In entire model, have one it is unidirectional
The information flow of flowing is that hidden unit is reached from input unit, and the information flow of another at the same time one-way flow is from hiding list
Member reaches output unit；Meanwhile the parameter learning of recurrent neural network is as traditional neural network algorithm, be equally using
Back-propagation algorithm, but if recurrent neural network is carried out network expansion, then parameter of the input layer to hidden layer, hidden layer
Between parameter, the data of hidden layer to output layer be shared, and using in gradient descent algorithm, the output of each step
The network currently walked is not only relied on, and also relies on the state of several step networks in front.
10. the oil field prospecting reserve forecasting method according to claim 1 based on recurrent neural network, which is characterized in that
In steps of 5, in forecast year, enterprise formulates production plan, the budget of workload, investment comprising investment, by investment
Input data of the budget of workload, investment as recurrent neural networks model, by recurrent neural network according to basic number
Accordingly and investment workload, the value of investment, pass through the propagated forward process of recurrent neural network network, prediction generate year
Predict the numerical value of proved reserves.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810855341.5A CN109002927A (en) | 2018-07-30 | 2018-07-30 | Oil field prospecting reserve forecasting method based on recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810855341.5A CN109002927A (en) | 2018-07-30 | 2018-07-30 | Oil field prospecting reserve forecasting method based on recurrent neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109002927A true CN109002927A (en) | 2018-12-14 |
Family
ID=64597898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810855341.5A Pending CN109002927A (en) | 2018-07-30 | 2018-07-30 | Oil field prospecting reserve forecasting method based on recurrent neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109002927A (en) |
Cited By (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 |
| CN110389948A (en) * | 2019-07-19 | 2019-10-29 | 南京工业大学 | A kind of tail oil prediction technique of the hydrocracking unit based on data-driven |
| CN110866913A (en) * | 2019-11-21 | 2020-03-06 | 桂林电子科技大学 | Deep recursion cardiovascular image display method |
| CN112785451A (en) * | 2021-01-07 | 2021-05-11 | 中国石油天然气股份有限公司 | Quantitative analysis method and device for oil and gas exploration effect |
| CN113379536A (en) * | 2021-06-29 | 2021-09-10 | 百维金科(上海)信息科技有限公司 | Default probability prediction method for optimizing recurrent neural network based on gravity search algorithm |
| CN115506783A (en) * | 2021-06-21 | 2022-12-23 | 中国石油化工股份有限公司 | Lithology identification method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140334262A1 (en) * | 2013-04-22 | 2014-11-13 | Robert H. Brune | Method and Apparatus for Active Seismic Shear Wave Monitoring of Hydro-Fracturing of Oil and Gas Reservoirs Using Arrays of Multi-Component Sensors and Controlled Seismic Sources |
| CN104951842A (en) * | 2014-03-27 | 2015-09-30 | 中国石油化工股份有限公司 | Novel method for predicting oil field output |
| CN105095986A (en) * | 2015-06-23 | 2015-11-25 | 中国石油天然气股份有限公司 | Multi-layer oil reservoir overall yield prediction method |
| CN105868508A (en) * | 2016-04-28 | 2016-08-17 | 西南石油大学 | Productivity quantitative prediction method based on gas logging information |
-
2018
- 2018-07-30 CN CN201810855341.5A patent/CN109002927A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140334262A1 (en) * | 2013-04-22 | 2014-11-13 | Robert H. Brune | Method and Apparatus for Active Seismic Shear Wave Monitoring of Hydro-Fracturing of Oil and Gas Reservoirs Using Arrays of Multi-Component Sensors and Controlled Seismic Sources |
| CN104951842A (en) * | 2014-03-27 | 2015-09-30 | 中国石油化工股份有限公司 | Novel method for predicting oil field output |
| CN105095986A (en) * | 2015-06-23 | 2015-11-25 | 中国石油天然气股份有限公司 | Multi-layer oil reservoir overall yield prediction method |
| CN105868508A (en) * | 2016-04-28 | 2016-08-17 | 西南石油大学 | Productivity quantitative prediction method based on gas logging information |
Non-Patent Citations (3)
| Title |
|---|
| ALEX GRAVES: "Supervised Sequence Labelling with Recurrent Neural Networks", 《SPRINGER》 * |
| J.SUN ET AL: "Comparison of Decline Curve Analysis DCA with Recursive Neural Networks RNN for Production Forecast of Multiple Wells", 《SPE》 * |
| 寂静的天空: "TensorFlow(十一)：递归神经网络（RNN与LSTM）", 《HTTPS：//WWW.CNBLOGS.COM/FELIXWANG2/P/9190656.HTML》 * |
Cited By (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 |
| CN110389948A (en) * | 2019-07-19 | 2019-10-29 | 南京工业大学 | A kind of tail oil prediction technique of the hydrocracking unit based on data-driven |
| CN110866913A (en) * | 2019-11-21 | 2020-03-06 | 桂林电子科技大学 | Deep recursion cardiovascular image display method |
| CN112785451A (en) * | 2021-01-07 | 2021-05-11 | 中国石油天然气股份有限公司 | Quantitative analysis method and device for oil and gas exploration effect |
| CN115506783A (en) * | 2021-06-21 | 2022-12-23 | 中国石油化工股份有限公司 | Lithology identification method |
| CN113379536A (en) * | 2021-06-29 | 2021-09-10 | 百维金科(上海)信息科技有限公司 | Default probability prediction method for optimizing recurrent neural network based on gravity search algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109002927A (en) | Oil field prospecting reserve forecasting method based on recurrent neural network | |
| Beven | How to make advances in hydrological modelling | |
| RU2573746C2 (en) | Systems and methods for well performance forecasting | |
| CN109800863B (en) | Logging phase identification method based on fuzzy theory and neural network | |
| CN108460134A (en) | The text subject disaggregated model and sorting technique of transfer learning are integrated based on multi-source domain | |
| US8090538B2 (en) | System and method for interpretation of well data | |
| Zhang et al. | Bilateral sensitivity analysis: a better understanding of a neural network | |
| CN106529667A (en) | Logging facies identification and analysis method based on fuzzy depth learning in big data environment | |
| CN105760673A (en) | Fluvial facies reservoir earthquake sensitive parameter template analysis method | |
| CN107704970A (en) | A kind of Demand-side load forecasting method based on Spark | |
| CN110428082A (en) | Water quality prediction method based on attention neural network | |
| CN112419711B (en) | Closed parking lot parking demand prediction method based on improved GMDH algorithm | |
| CN106529185A (en) | Historic building displacement combined prediction method and system | |
| Bai et al. | Efficient and data-driven prediction of water breakthrough in subsurface systems using deep long short-term memory machine learning | |
| CN107945534A (en) | A kind of special bus method for predicting based on GMDH neutral nets | |
| CN114944053A (en) | Traffic flow prediction method based on spatio-temporal hypergraph neural network | |
| CN110110372A (en) | A kind of user's timing behavior automatic segmentation prediction technique | |
| KR20140137210A (en) | computing system for well placement optimization developed by SA/ANN and well placement optimization method using Thereof | |
| Hong et al. | Value-of-information for model parameter updating through history matching | |
| Caihong et al. | Real-time flood classification forecasting based on k-means++ clustering and neural network | |
| CN113111256A (en) | Production work order recommendation method based on depth knowledge map | |
| Ju et al. | Hydrologic simulations with artificial neural networks | |
| CN104102837A (en) | Incremental learning integrated algorithm for soft measurement modeling | |
| Hutahaean | Multi-objective methods for history matching, uncertainty prediction and optimisation in reservoir modelling | |
| CN107229944A (en) | Semi-supervised active identification method based on cognitive information particle |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181214 |
|
| RJ01 | Rejection of invention patent application after publication |