CN116644844A - Stratum pressure prediction method based on neural network time sequence - Google Patents

Stratum pressure prediction method based on neural network time sequence Download PDF

Info

Publication number
CN116644844A
CN116644844A CN202310570384.XA CN202310570384A CN116644844A CN 116644844 A CN116644844 A CN 116644844A CN 202310570384 A CN202310570384 A CN 202310570384A CN 116644844 A CN116644844 A CN 116644844A
Authority
CN
China
Prior art keywords
model
neural network
network time
parameters
output
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310570384.XA
Other languages
Chinese (zh)
Inventor
王子振
陈冠霖
张锐
周卫东
王成文
王翔
李罗鹏
陈修平
李明忠
宋文宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Petroleum East China
Original Assignee
China University of Petroleum East China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China University of Petroleum East China filed Critical China University of Petroleum East China
Priority to CN202310570384.XA priority Critical patent/CN116644844A/en
Publication of CN116644844A publication Critical patent/CN116644844A/en
Pending legal-status Critical Current

Links

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"
    • 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
    • E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
    • 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
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mining & Mineral Resources (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Geology (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Agronomy & Crop Science (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Development Economics (AREA)
  • Animal Husbandry (AREA)
  • Marine Sciences & Fisheries (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Environmental & Geological Engineering (AREA)
  • Fluid Mechanics (AREA)
  • Game Theory and Decision Science (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geochemistry & Mineralogy (AREA)

Abstract

The application belongs to the field of petroleum exploration and development engineering, and particularly relates to a stratum pressure prediction method based on a neural network time sequence. The method mainly comprises the following steps: collecting field logging data and carrying out noise reduction filtering treatment; selecting proper model input parameters; selecting a training set to construct a proper neural network time sequence model, and optimizing the number of neurons and time lag parameters; and selecting the predicted formation pressure of the new well to test the generalization performance of the method. The neural network time sequence (NARX) model used in the application considers the depth sequence, can utilize the data of the upper stratum to predict the lower stratum, effectively avoids the problem of poor generalization of the conventional neural network model and other machine learning models, establishes the data internal relation between logging data and stratum pressure, and excavates the historical information in the depth sequence, thereby solving the problems of low prediction precision, poor generalization and no consideration of the upper stratum data in the conventional stratum pressure prediction method.

Description

Stratum pressure prediction method based on neural network time sequence
Technical Field
The application belongs to the field of petroleum exploration and development engineering, and particularly relates to a stratum pressure prediction method based on a neural network time sequence.
Background
The formation pressure is an important parameter in the oil and gas exploration and development process, and the high-precision prediction result is the guarantee of safe and efficient drilling construction. The conventional stratum pressure prediction method mainly comprises an Eaton method, a Bowers method and the like, the conventional methods often need to establish a normal trend line, the normal trend line is established without unified standards and is generally established according to experience of researchers, so that non-professional persons are difficult to apply, meanwhile, the conventional prediction method does not consider the change of stratum properties, ignores the role of a depth sequence in stratum pressure prediction, and is difficult to meet the high-precision requirement on stratum pressure in deep water deep well drilling engineering. With the development of logging while drilling and logging while drilling technology, more and more data can be used for formation pressure prediction, the mass data are well utilized by a nonlinear modeling means in artificial intelligence, a new thought is provided for solving the problem of poor application effect caused by too simple linear mathematical model in the conventional method, but the current research is limited by a plurality of formation pressure influence factors, large formation property change, complex pressure anomaly cause and the like, and no good progress is made.
Disclosure of Invention
The application aims to provide a stratum pressure prediction method based on a neural network time sequence, which solves the problems that the existing stratum pressure prediction method is low in prediction precision and poor in generalization and upper stratum data is not considered.
In order to achieve the aim of the application, the application adopts the following technical scheme: a stratum pressure prediction method based on a neural network time sequence comprises the following prediction steps:
s1, collecting on-site drilling construction logging data;
s2, performing noise reduction and filtering treatment on logging data collected on site;
s3, dividing the data subjected to noise reduction filtering processing into a training set, a verification set and a test set, and determining model input and model output;
s4, determining a model error evaluation method;
s5, constructing a neural network time sequence model by using the training set, and optimizing model super parameters by combining the error evaluation method in the S4 and the verification set;
the construction and optimization process of the model is as follows:
s5-1, model structural design: determining the structure of an NARX model, wherein the NARX model comprises an input layer, a hidden layer and an output layer;
s5-2, initializing parameters: initializing model parameters, wherein the parameters are weights and biases;
s5-3. Forward propagation: the model input parameters of the training set are sequentially calculated to output values of a hidden layer and an output layer through a forward propagation process of the model, and prediction output is obtained;
s5-4, loss calculation: calculating a loss function using a difference between the predicted output and the target output;
s5-5. Counter propagation: calculating the gradient of the model parameters (weight and bias) of the loss function through a back propagation algorithm, and adjusting the value of the model parameters according to the direction and the size of the gradient so as to gradually reduce the loss function;
s5-6, parameter optimization: updating parameters of the model using an optimization algorithm (e.g., random gradient descent, adam, etc.) in combination with the validation set to minimize the loss function;
s5-7, repeating the iteration: repeating the steps S5-3 to S5-6 until reaching a predefined stopping condition (such as reaching a maximum iteration number or small change of a loss function) to obtain a trained model;
s6, testing the trained model according to the test set;
s7, carrying out stratum pressure prediction by using logging data of a new well.
The logging data includes depth, weight on bit, hook load, torque, riser pressure, displacement, rate of penetration, formation pressure gradient.
In the step S2, a gaussian moving average method is used to perform noise reduction filtering.
The training set, the verification set and the test set in the step S3 are in a proportion of 75 percent: 15%:15, model inputs are weight on bit, hook load, torque, riser pressure, displacement and rate of penetration; the model output is the stratum pressure gradient, and the 6 parameters are parameters which can be obtained in real time in logging data, so that the timeliness of stratum pressure prediction is improved, the change of stratum property and pore fluid property is reflected, and the defect that the stratum property is not changed in a conventional stratum pressure model is overcome.
The specific formula of the error evaluation method in the step S4 is as follows:
；
wherein:
n-total number of samples;
y i sample true value, g/cm 3 ；
Sample predictive value, g/cm 3 ；
Average of all samples, g/cm 3 ；
RMSERoot mean square error, g/cm 3 ；
R 2 -goodness of fit.
The neural network time sequence model in the step S5 is an NARX network model, is a nonlinear autoregressive neural network and has the characteristic of a dynamic neural network, namely, the output of the network depends on the current external input and also depends on the historical output result, so that compared with a common neural network model, the NARX network has the advantages of memory and better association capability, stronger self-adaptation capability, capability of better utilizing the historical output result and accordance with the aim of carrying out stratum pressure prediction and updating by utilizing the existing result. The NARX network model can be expressed as:
；
in the method, in the process of the application,y(t) In order to output the variable(s),x(t) In order to input the variable(s),y(t-1)，y(t-2)，y(t-n) As a result of the history of the output variables,x(t-1)，x(t-2)，x(t-n) As a result of the history of the input variables,nis the time lag size (the past time step size in the NARX model).
The hidden layer activation function and the output layer activation function of the NARX network model are respectively:
；
wherein:
x-model input;
f 1 (x)——activating a function by the hidden layer;
f 2 (x) -output layer activation function;
e-natural constant.
The super parameters to be optimized in the step S5 are time lag size and neuron number.
In the step S7, logging data of a brand new well is selected to be substituted into the model for prediction, and the generalization capability of the model is checked.
The beneficial effects of the application are concentrated in that:
(1) The method utilizes the advantage of real-time availability of logging data, combines a neural network time sequence model, can calculate the formation pressure in real time, is beneficial to the real-time prediction work of the formation pressure in the drilling process, and ensures the safe and efficient drilling work.
(2) The prediction method is not limited by a normal compaction curve, overcomes the defects of lower precision and dependence on subjective judgment of researchers of the traditional method, and has wider application prospect.
(3) According to the application, the advantage of a neural network time sequence (NARX) model is utilized, the change trend is extracted from the historical value, the data of the upper stratum is fully utilized to correct the prediction result, the model precision is greatly improved, the model fitting goodness is more than 95%, and the problem of poor application effect of the conventional machine learning method caused by large stratum property change is solved.
Drawings
FIG. 1 is a predictive flow diagram of the present application;
FIG. 2 is a comparison of the data noise reduction of the present application;
FIG. 3 is a graph of formation pressure predictions in accordance with the present application
Detailed Description
In order to make the technical solution of the present application better understood by those skilled in the art, the present application will be further described in detail with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, a formation pressure prediction method based on a neural network time sequence includes the following prediction steps:
1. existing in situ logging data is collected, and is selected for modeling as is typical, such data being available from logging while drilling tools in real-world applications. The logging data used by the application comprises depth, hook load, weight on bit, torque, displacement, mechanical drilling speed, riser pressure and stratum pressure gradient, and the logging data used in the case is logging data of one well drilled X1 well in the south China sea, which is 2655 lines in total.
2. The existing logging data is subjected to noise reduction and filtering treatment, and the depth is a real parameter, so that the stratum pressure gradient is a parameter to be predicted, and therefore, the noise reduction and filtering treatment is only carried out on six variables, namely the hook load, the drilling pressure, the torque, the displacement, the mechanical drilling speed and the riser pressure, and the parameters are possibly influenced by the precision and the sampling frequency of a signal collector during on-site acquisition, so that the fluctuation of a sampling curve is larger, and the modeling precision of a model is influenced. The results before and after noise reduction and filtering of different parameters are shown in fig. 2.
3. Determining 6 variables of hook load, weight on bit, torque, displacement, mechanical parameters and riser pressure as model input, wherein the 6 parameters are parameters which can be obtained in real time in logging data, are beneficial to improving timeliness of formation pressure prediction, are beneficial to reflecting changes of formation properties and pore fluid properties, overcome the defect that the formation properties are not changed in a conventional formation pressure model, output by taking formation pressure gradient variables as models, and the ratio of a training set, a verification set and a test set is 70%:15%:15%, randomly sampling from the well data to achieve the aim of improving the generalization performance of the model.
4. In this embodiment, two error evaluation methods of the evaluation model are respectively:
；
wherein:
n-total number of samples;
y i sample true value, g/cm 3 ；
Sample predictive value, g/cm 3 ；
Average of all samples, g/cm 3 ；
RMSERoot mean square error, g/cm 3 ；
R 2 -goodness of fit.
5. The neural network time series model in this embodiment is modeled by a training set, validated by set super parameter optimization and combined with the error evaluation method in S4, and the time lag of use is determined to be 5 and the number of neurons in a hidden layer is determined to be 10 by considering the comprehensive computer modeling time, so that an optimal model is obtained.
6. In this embodiment, a model test is performed using reserved 15% data as a test set, with a model RMSE of 1.22×10 -3 g/cm 3 At this time, the model fitting goodness is over 90%, but at this time, all the test set data and the training set are in the same well, and the on-site practical application effect is considered, if the model needs to be verified whether the model has good generalization, the model test is also needed to be performed by using new well data.
7. In this embodiment, an X2 well with a distance close to that of the X1 well and similar formation properties is selected for model secondary test, the well has 2399 rows of data, and the result is shown in FIG. 3, and the model error RMSE in this embodiment is 1.89×10 -3 g/cm 3 Goodness of fit R 2 = 0.9986, which demonstrates that the model has higher accuracy of formation pressure prediction during drilling of similar formations.
Based on the traditional stratum pressure, the application considers the depth sequence of logging data, and uses the neural network time sequence to provide a new stratum pressure prediction method. By utilizing the characteristic that logging data can be obtained in real time, the relation between historical data and future data is mined, and the formation pressure prediction accuracy is improved.
Compared with the traditional method, the prediction method is not limited by a normal compaction curve, overcomes the defects of lower precision and dependence on subjective judgment of researchers of the traditional method, and has wider application prospect.
According to the application, the advantage of a neural network time sequence (NARX) model is utilized, the change trend is extracted from the historical value, the data of the upper stratum is fully utilized to correct the prediction result, the model precision is greatly improved, the model fitting goodness is more than 95%, and the problem of poor application effect of the conventional machine learning method caused by large stratum property change is solved.
It should be noted that, for simplicity of description, the foregoing method embodiments are all expressed as a series of action combinations, but it should be understood by those skilled in the art that the present application is not limited by the order of action described, as some steps may be performed in other order or simultaneously according to the present application. Further, it should be understood by those skilled in the art that the embodiments described in the specification are all preferred embodiments, and the acts and elements referred to are not necessarily required for the present application.

Claims (8)

1. The stratum pressure prediction method based on the neural network time sequence is characterized by comprising the following steps of:
s1, collecting on-site drilling construction logging data;
s2, performing noise reduction and filtering treatment on logging data collected on site;
s3, dividing the data subjected to noise reduction filtering processing into a training set, a verification set and a test set, and determining model input and model output;
s4, determining a model error evaluation method;
s5, constructing a neural network time sequence model by using the training set, and optimizing model super parameters by combining the error evaluation method in the S4 and the verification set;
the construction and optimization process of the model is as follows:
s5-1, model structural design: determining the structure of an NARX model, wherein the NARX model comprises an input layer, a hidden layer and an output layer;
s5-2, initializing parameters: initializing model parameters, wherein the parameters are weights and biases;
s5-3. Forward propagation: the model input parameters of the training set are sequentially calculated to output values of a hidden layer and an output layer through a forward propagation process of the model, and prediction output is obtained;
s5-4, loss calculation: calculating a loss function using a difference between the predicted output and the target output;
s5-5. Counter propagation: calculating the gradient of the loss function to the model parameters through a back propagation algorithm, and adjusting the values of the model parameters according to the direction and the size of the gradient so as to gradually reduce the loss function;
s5-6, parameter optimization: updating parameters of the model using an optimization algorithm in combination with the validation set to minimize the loss function;
s5-7, repeating the iteration: repeating the steps S5-3 to S5-6 until a predefined stopping condition is reached, so as to obtain a trained model;
s6, testing the trained model according to the test set;
s7, carrying out stratum pressure prediction by using logging data of a new well.
2. The neural network time series based formation pressure prediction method of claim 1, wherein the logging data comprises depth, weight on bit, hook load, torque, riser pressure, displacement, rate of penetration, formation pressure gradient.
3. The method for predicting formation pressure based on neural network time series according to claim 1, wherein in the step S2, a gaussian moving average method is used for noise reduction filtering.
4. The method for predicting formation pressure based on neural network time series according to claim 1, wherein the training set, the validation set and the test set in the step S3 are in a ratio of 75%:15%:15%; model inputs are weight on bit, hook load, torque, riser pressure, displacement, and rate of penetration; the model output is the formation pressure gradient.
5. The method for predicting formation pressure based on neural network time series according to claim 1, wherein the specific formula of the error evaluation method in step S4 is as follows:
；
wherein:
n-total number of samples;
y i sample true value, g/cm 3 ；
Sample predictive value, g/cm 3 ；
Average of all samples, g/cm 3 ；
RMSERoot mean square error, g/cm 3 ；
R 2 -goodness of fit.
6. The method for predicting formation pressure based on neural network time series according to claim 1, wherein the neural network time series model in step S5 is a NARX network model, expressed as:
；
in the method, in the process of the application,y(t) In order to output the variable(s),x(t) In order to input the variable(s),y(t-1)，y(t-2)，y(t-n) For calendarThe history is output of a variable which,x(t-1)，x(t-2)，x(t-n) As a result of the history of the input variables,nthe time lag is the size.
7. The method for predicting formation pressure based on neural network time series as claimed in claim 6, wherein the hidden layer activation function and the output layer activation function of the NARX network model are respectively:
；
wherein:
x-model input;
f 1 (x) -a hidden layer activation function;
f 2 (x) -output layer activation function;
e-natural constant.
8. The method according to claim 1, wherein the super-parameters in the step S5 are time lag and neuron number.
CN202310570384.XA 2023-05-19 2023-05-19 Stratum pressure prediction method based on neural network time sequence Pending CN116644844A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310570384.XA CN116644844A (en) 2023-05-19 2023-05-19 Stratum pressure prediction method based on neural network time sequence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310570384.XA CN116644844A (en) 2023-05-19 2023-05-19 Stratum pressure prediction method based on neural network time sequence

Publications (1)

Publication Number Publication Date
CN116644844A true CN116644844A (en) 2023-08-25

Family

ID=87618124

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310570384.XA Pending CN116644844A (en) 2023-05-19 2023-05-19 Stratum pressure prediction method based on neural network time sequence

Country Status (1)

Country Link
CN (1) CN116644844A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117952013A (en) * 2024-02-01 2024-04-30 深圳市威鹏建设科技有限公司 Geotechnical engineering structural mode prediction analysis method and system

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117952013A (en) * 2024-02-01 2024-04-30 深圳市威鹏建设科技有限公司 Geotechnical engineering structural mode prediction analysis method and system
CN117952013B (en) * 2024-02-01 2024-08-13 深圳市威鹏建设科技有限公司 Geotechnical engineering structural mode prediction analysis method and system

Similar Documents

Publication Publication Date Title
CN110807557B (en) BP neural network-based drilling rate prediction method and BP neural network-based particle swarm optimization method
US20210350208A1 (en) Method and device for predicting production performance of oil reservoir
CN109543828A (en) A kind of intake profile prediction technique based under condition of small sample
CN115049173B (en) Deep learning and Eaton method coupling driving stratum pore pressure prediction method
CN116384554A (en) Method and device for predicting mechanical drilling speed, electronic equipment and computer storage medium
CN116644844A (en) Stratum pressure prediction method based on neural network time sequence
CN114723095A (en) Missing well logging curve prediction method and device
CN115238861B (en) Safe drilling fluid tightness determining method based on well wall collapse degree constraint
CN115099406A (en) Stratum pressure inversion method and device based on multivariate time sequence
CN114004100B (en) Oil reservoir assisted history fitting and optimization simulation method
CN112016766A (en) Oil and gas well drilling overflow and leakage early warning method based on long-term and short-term memory network
CN115860197A (en) Data-driven coal bed gas yield prediction method and system
CN115059448A (en) Stratum pressure monitoring method based on deep learning algorithm
CN113494286B (en) Intelligent dynamic prediction method and system for drilling speed in geological drilling process
CN115481565A (en) Earth pressure balance shield tunneling parameter prediction method based on LSTM and ant colony algorithm
CN116976146B (en) Fracturing well yield prediction method and system coupled with physical driving and data driving
CN116911216B (en) Reservoir oil well productivity factor assessment and prediction method
CN110486008B (en) Parameter interpretation method and system for radial composite oil reservoir
CN110486009B (en) Automatic parameter reverse solving method and system for infinite stratum
CN111749675A (en) Stratum drillability prediction method and system based on cascade model algorithm
CN111751878A (en) Method and device for predicting transverse wave velocity
Mnati et al. Prediction of penetration rate and cost with artificial neural network for alhafaya oil field
CN115618750A (en) Underground oil-water seepage agent model based on coupling neural network
CN111625925B (en) Ternary combination flooding injection-production optimization method based on chromatographic separation
CN114547748B (en) Method and system for optimally designing construction scheme of horizontal well staged fracturing technology

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