CN111638249B - Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation - Google Patents
Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation Download PDFInfo
- Publication number
- CN111638249B CN111638249B CN202010481478.6A CN202010481478A CN111638249B CN 111638249 B CN111638249 B CN 111638249B CN 202010481478 A CN202010481478 A CN 202010481478A CN 111638249 B CN111638249 B CN 111638249B
- Authority
- CN
- China
- Prior art keywords
- layer
- convolution
- sample
- water content
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 84
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 239000003129 oil well Substances 0.000 title claims abstract description 12
- 239000012530 fluid Substances 0.000 claims abstract description 51
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000005514 two-phase flow Effects 0.000 claims abstract description 16
- 238000005259 measurement Methods 0.000 claims abstract description 14
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000012360 testing method Methods 0.000 claims abstract description 6
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 24
- 238000011176 pooling Methods 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 13
- 210000002569 neuron Anatomy 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 4
- 230000002123 temporal effect Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 238000012935 Averaging Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims description 2
- 230000008859 change Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000003556 assay Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/02—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance
- G01N27/04—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance by investigating resistance
- G01N27/06—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating impedance by investigating resistance of a liquid
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Chemical & Material Sciences (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A water content measuring method based on deep learning and application thereof in oil well exploitation comprise: collecting fluid data of different working conditions in a wellhead oil and gas pipeline by using a four-sector conductivity sensor, acquiring a label, and uploading the label to an upper computer for storage; constructing a data set, specifically, after preprocessing fluid data, obtaining a sample from the fluid data by using a sliding window with overlap, adding a corresponding label to the sample, taking 80% of the sample as a training set, 10% of the sample as a verification set, and taking the remaining 10% of the sample as a test set; constructing a convolutional neural network model for completing the feature extraction and the water content value prediction of the fluid sample; and optimizing, training and adjusting parameters of the convolutional neural network model to obtain the optimal convolutional neural network model under the current architecture, and completing parameter measurement. The invention can accurately measure and store the water content value of the oil-water two-phase flow under the wellhead environment in real time, and improves the accuracy of water content measurement.
Description
Technical Field
The invention relates to a method for measuring water content. In particular to a water content measuring method based on deep learning and application thereof in oil well exploitation.
Background
The oil-water two-phase flow is widely existed in the industrial fields of oil and gas exploitation, transportation, processing and the like. In an oil-water two-phase flow system, due to the difference of properties such as density, viscosity and the like, the distribution of two-phase media in a pipeline has changes in both space and time, and a specific flow form, namely a flow pattern, is formed. In the oil and gas exploitation process, along with the change of the flow pattern in the two-phase flow system, the whole flow information is continuously changed, so that the key flow parameters (phase flow rate, phase content and the like) are also in violent change, the difficulty is brought to the wellhead content measurement of the oil and gas, and the further processes of transportation, storage and the like are influenced. Currently, the mainstream method for determining the water content of the wellhead is to perform manual assay on a sampling product so as to obtain the water content of the current sample. However, this method has large error, low precision and large time delay. A model of fluid signals and flow parameters in a wellhead pipeline is built, the model can be used for measuring the flow parameters of different fluid signals, the model is continuously optimized from two angles of fluid signal acquisition and model optimization, and the performance of the model is improved. Traditional annular conductivity sensors, double helix capacitance sensors and the like only contain single-channel fluid information and cannot accurately reflect flow structures. The distributed conductivity sensor design can acquire richer spatial information, and the flow state is reflected more accurately through multi-channel signals, so that technical support is provided for accurate measurement of flow parameters.
The deep learning method is an important technical route for realizing artificial intelligence under the current technological background. Through the simulation of a human brain processing mechanism, the deep learning method utilizes the integration of artificial neurons to complete large-scale nonlinear calculation. In recent years, with the solution of corresponding optimization problems, the deep learning method obtains the accuracy rate close to or even exceeding that of human brain recognition in a plurality of fields, and the development of artificial intelligence is powerfully promoted. The convolutional neural network is a branch which is widely applied, and has excellent performance in the fields of image segmentation, natural language processing, signal analysis and the like by virtue of the advantages of local perception, parameter sharing and multi-core processing. Meanwhile, with the continuous improvement of the convolutional neural network, the processing performance of the convolutional neural network is also continuously improved. Compared with the traditional analysis method, the convolutional neural network can acquire representative characteristics through analysis of a large amount of data, and is beneficial to analysis of new data. The network has a plurality of feature extraction processes, so that the nonlinear features in the signals can be fully discovered through the action of the nonlinear function, and the network has unique advantages in the processing of complex signals.
Disclosure of Invention
The invention aims to solve the technical problem of providing a deep learning-based water content measuring method capable of solving the problem of measuring the water content of oil-water two-phase flow and application thereof in oil well exploitation
The technical scheme adopted by the invention is as follows: a moisture content measuring method based on deep learning comprises the following steps:
1) collecting fluid data of different working conditions in a wellhead oil and gas pipeline by using a four-sector conductivity sensor, acquiring a label, and uploading the label to an upper computer for storage;
2) constructing a data set, specifically, after preprocessing fluid data, obtaining a sample from the fluid data by using a sliding window with overlap, adding a corresponding label to the sample, taking 80% of the sample as a training set, 10% of the sample as a verification set, and taking the remaining 10% of the sample as a test set;
3) constructing a convolutional neural network model for completing the feature extraction and the water content value prediction of the fluid sample;
4) and optimizing, training and adjusting parameters of the convolutional neural network model to obtain the optimal convolutional neural network model under the current architecture, and completing parameter measurement.
An application of a water content measuring method based on deep learning in oil well exploitation comprises the following steps:
(1) starting a four-sector conductivity sensor, measuring fluid data of oil-water two-phase flow in a wellhead oil-gas pipeline, uploading the fluid data to an upper computer, and obtaining a sample at the current moment according to the water content measuring method based on deep learning;
(2) taking the current time sample as the input of a convolutional neural network model, and obtaining the water content output of the current sample through the calculation and analysis of the convolutional neural network model;
(3) the water content output is associated with the corresponding sample and stored in an upper computer as the current processing information;
(4) and (4) repeating the steps (2) to (3), continuously obtaining the water content value corresponding to the fluid sample in real time, storing the water content value as the state information of the current wellhead oil and gas pipeline until the four-sector conductivity sensor is closed, and analyzing historical data to obtain the total water content in the measuring time.
According to the deep learning-based water content measuring method and the application thereof in oil well exploitation, the oil-water two-phase flow water content measuring model is combined with the four-sector conductivity sensor, and the water content measuring system is constructed together with the storage module and the display module, so that the water content value of the oil-water two-phase flow under a wellhead environment can be accurately measured in real time and stored, the accuracy of water content measurement is improved, and a beneficial thought is provided for solving the problem of water content measurement of the oil-water two-phase flow.
Drawings
FIG. 1 is a flow chart of the method for measuring moisture content based on deep learning of the present invention;
FIG. 2 is a schematic of a four sector conductivity sensor of the present invention for acquiring multi-channel fluid signals;
in the figures, 1-4: an excitation electrode; 5-8: a signal acquisition electrode;
FIG. 3 is a flow chart of the convolutional neural network model construction for measuring water cut of the present invention;
fig. 4 is a convolutional neural network architecture based on a four-sector fluid signal of the present invention.
Detailed Description
The deep learning-based water cut measuring system and the application thereof in oil well exploitation are described in detail below with reference to the embodiments and the accompanying drawings.
The water content measuring method based on deep learning uses the deep learning method of the convolutional neural network to train on four-sector experimental data of vertical oil-water two-phase flow, obtains a model capable of accurately measuring the water content in real time, is deployed in a water content measuring system, solves the problems of complex change and difficult prediction of fluid signals to a certain extent, and can apply the measuring system to the actual wellhead environment for field measurement.
As shown in FIG. 1, the method for measuring moisture content based on deep learning of the present invention comprises the following steps:
1) collecting fluid data of different working conditions in a wellhead oil and gas pipeline by using a four-sector conductivity sensor shown in figure 2, acquiring a label, and uploading the label to an upper computer for storage; the method comprises the following steps:
(1) the four-sector conductivity sensor arranged on the transparent pipe section is connected with a wellhead oil and gas acquisition pipeline through a flange, fluid data of oil-water two-phase flow under different working conditions in the wellhead oil and gas pipeline are acquired, the measurement time is 20min, the sampling frequency of the sensor is 2000Hz, and the data are uploaded to an upper computer for storage;
(2) collecting an oil-water sample flowing through a transparent pipe section in the working time of the four-sector conductivity sensor to obtain a water content label of the oil-water sample, determining the overall range of the water content according to the characteristics of an oil well, taking 3% of the water content in the overall range as a step point, and selecting fluid data for subsequent model training.
2) Constructing a data set, specifically, after preprocessing fluid data, obtaining a sample from the fluid data by using a sliding window with overlap, adding a corresponding label to the sample, taking 80% of the sample as a training set, 10% of the sample as a verification set, and taking the remaining 10% of the sample as a test set; the method comprises the following steps:
(1) fluid data is pre-processed, with the following formula:
wherein,
is the ith fluid data of the jth sector,
and
respectively the mean and standard deviation of the j-th sector fluid data,
the ith fluid data of the preprocessed jth sector;
(2) intercepting fluid data of each working condition as a single sample through a sliding window with overlap, wherein the label of the sample refers to a water content label, each group of water content labels is regarded as one type, obtaining a class label of the sample, and setting the length of the sliding window as M and the overlap rate asPoThen the single sample is a two-dimensional matrix of M × 4, and is obtained from the fluid data with length L
A sample of, wherein,
the representative data is rounded down, and N samples of the fluid data under all working conditions are obtained;
in the embodiment of the present invention, the length of the sliding window used is 3000, the overlap ratio is 35%, then a single sample is a two-dimensional matrix of 3000 × 4, 1230 samples can be cut out from the signal with the length of L, and a total of 13530 data samples are obtained from 11 operating conditions.
(3) Randomly using 80% of N samples as training set, 10% as verification set and the rest 10% as test set.
3) Constructing a convolutional neural network model for completing the feature extraction and the water content value prediction of the fluid sample;
the convolutional neural network model is used for extracting spatial information and time domain characteristics in two-phase flow data and measuring the water content through a multi-task output module, as shown in fig. 3; the convolutional neural network model has 19 layers as shown in FIG. 4, and comprises 4 convolutional modules and 1 multitask output module; the layers 1-4 are first convolution modules, the layers 5-8 are second convolution modules, the layers 9-12 are third convolution modules, the layers 13-15 are fourth convolution modules, and the layers 16-19 are multitask output modules. Wherein,
A) the first convolution module, the second convolution module and the third convolution module have the same structure and respectively comprise: a temporal convolution layer, a spatial convolution layer, an average pooling layer, and a discard layer; the fourth convolution module comprises the following components which are connected in sequence: a temporal convolution layer, a spatial convolution layer, and a discard layer; wherein,
(1) the time convolution layer is used for filtering the four-channel samples along the time dimension and extracting time domain characteristics; the time convolution layer uses
Each convolution kernel is corresponding to a first convolution module, a second convolution module, a third convolution module and a fourth convolution module, wherein k is 1,2,3 and 4, and different feature maps are obtained after each convolution kernel and the input sample are subjected to convolution operation to obtain the different feature maps
A feature map; the formula for this layer convolution is as follows:
wherein,
is an input to the time convolution layer,
is a characteristic map of the time convolution layer output, fTIs the activation function of the time convolution layer, kTIndex of points in the convolution kernel for time convolution layer, GTFeature dimension, g, input for time convolution layerTThe feature index input for the time convolution layer,
and
respectively, the p-th in the time convolution layerTA convolution kernel and an offset, and,
the size of the convolution kernel is 1,2,3,4, which corresponds to the first convolution module and the second convolution module respectivelyA module, a third convolution module and a fourth convolution module, StTIs the step size of the time convolution layer convolution kernel, subTIs a subscript shorthand for time convolution layer input, mT,nTA location index of features extracted for the time convolution layer;
the boundary value is extended by complementing 0, and the size of each feature graph is IT/StT×STIn which IT,STIs the dimension of the time convolution layer input; carrying out batch normalization processing on the convolved features to accelerate convergence;
in the embodiment of the present invention, the step size of the convolution kernel used by the time convolution layer is 1, and the activation function is Relu. The time convolution layer in the first convolution module uses 20 convolution kernels, the size of the time convolution layer is 30, the time convolution layer in the second convolution module uses 20 convolution kernels, the size of the time convolution layer is 30, the time convolution layer in the third convolution module uses 20 convolution kernels, the size of the time convolution layer is 10, and the time convolution layer in the fourth convolution module uses 20 convolution kernels, the size of the time convolution layer is 10.
(2) The space convolution layer is used for integrating the samples of the four channels along the space dimension to obtain space information; the space convolution layer is used
A convolution kernel, wherein k is 1,2,3,4, respectively corresponding to the first convolution module, the second convolution module, the third convolution module and the fourth convolution module,
the convolution cores integrate the input feature maps to obtain a total
The characteristic diagram is as follows:
wherein,
is an input for the space convolution layer,
is a characteristic map of the space convolution layer output, fCIs the activation function of the space convolution layer, kCIndex of points in convolution kernel for spatial convolution layer, GCFeature dimension, g, input for space convolution layerCA feature index input for the spatial convolution layer,
and
respectively being the p-th in the space convolution layerCA convolution kernel and an offset, LeCAnd StCRespectively, the size and step size of the convolution kernel of the space convolution layer, subCSubscript abbreviation, m, of space convolution layer inputC,nCA location index of features extracted for the spatial convolution layer;
the input boundary value is extended by complementing 0, and the size of each feature map is IC/StC×SCIn which IC,SCThe dimension of the input dimension of the space convolution layer is used for carrying out batch normalization processing on the feature graph after convolution so as to accelerate convergence;
in the embodiment of the present invention, the size of the convolution kernel used by the spatial convolution layer is 4, the step size is 1, and the activation function is Relu. The space convolution layer in the first convolution module uses 20 convolution kernels, the space convolution layer in the second convolution module uses 20 convolution kernels, the space convolution layer in the third convolution module uses 20 convolution kernels, and the space convolution layer in the fourth convolution module uses 20 convolution kernels.
(3) The average pooling layer is used for selecting representative values in the pooling area to reduce characteristic dimensionality, removing redundant information and simplifying subsequent calculation, and a formula is described as follows:
wherein,
to average out the characteristics of the pooled layer output,
in order to average the inputs to the pooling layer,
for averaging the size of the pooling kernel, where k is 1,2,3, corresponding to the first, second and third convolution modules, gA,mA,nAIndex for average pooling layer characteristics, kAAn index of points in the average pooled kernel;
in the embodiment of the present invention, the pooling kernel size of the first convolution module is 5, the pooling kernel size of the second convolution module is 5, and the pooling kernel size of the third convolution module is 3.
(4) And a discarding layer, wherein a part of neurons and connections thereof are discarded randomly according to a discarding rate, so that the dense neural network model is forced to learn more global characteristics. In the present embodiment, the discard rate is set to 0.4.
B) The multitask output module comprises a flattening layer, a discarding layer and a full connection layer which are sequentially connected, wherein the output of the full connection layer is respectively connected with a numerical value full connection layer and an auxiliary full connection layer; wherein:
(1) the flattening layer is used for integrating the input multi-dimensional features into one-dimensional comprehensive information, so that the features which have important influence on the water content measurement can be conveniently obtained in the subsequent processing; the output dimension of the layer is IF×SF×NFIn which IF,SFIs the dimension of the size of the input of the flattening layer, NFIs a flat layer inputA quantity dimension of;
(2) the discarding layer discards part of neurons and connection thereof randomly according to a discarding rate, and the output dimension is N1 DHaving N of1 D＝IF×SF×NF；
(3) The full connection layer obtains brand new output characteristics by weighting and recombining the input characteristics; in the output characteristics, the effective characteristics are enhanced, and the formula is as follows:
wherein,
for the features of the full link layer output,
for input of full connection layer, fDFor the activation function of the fully connected layer,
the number of neurons input to the fully connected layer,
weight and offset, k, of the full connection layer, respectivelyDInput index, g, for the full connection layerDThe index is output for the full-link layer,
is the number of output neurons;
in the embodiment of the present invention, the activation function Relu of the fully-connected layer is 50 for the number of fully-connected layer output neurons.
(4) The numerical fully-connected layer is provided with 1 neuron, and the activation function of the layer is Sigmoid; the auxiliary full connection layer has Nu19And (4) a neuron, wherein the layer activation function is Softmax.
In the present example, the number of auxiliary full-link layer neurons was set to 11.
4) Optimizing, training and adjusting parameters of the convolutional neural network model to obtain the optimal convolutional neural network model under the current architecture, and completing parameter measurement; the method comprises the following steps:
(1) constructing a loss function between a label and an output of the convolutional neural network model, describing a difference between the label and the output by using the loss function, and guiding the optimization of the convolutional neural network model by taking the minimum difference as a principle; loss L of the convolutional neural network modeltotalComprises two parts, respectively: a) loss of numerically fully connected layers LpredSelecting a Mean Absolute Error (MAE) as a loss function; b) loss L of auxiliary full link layerclaSelecting a cross entropy loss function; loss LtotalThe expression is as follows:
Lcla＝-∑p(x)×log q(x)
Ltotal＝α×Lpred+β×Lcla
wherein, yi,
Respectively labeling and outputting the water content of the convolutional neural network model, wherein n is the number of samples; p (x), q (x) are the probability distributions of class labels and class outputs, respectively; α, β are the weights of the two-part loss function;
in the present embodiment, α is set to 10 and β is set to 0.8.
(2) Setting parameters of each layer in the convolutional neural network model, performing iterative training on the weight of the convolutional neural network model under the parameter combination by adopting a gradient descent method based on adam (adaptive motion estimation), and updating the weight of the network by a gradient back propagation method; during training, the learning rate is set to LrThe number of running steps is set to nepoch；
In the embodiment of the present invention, the learning rate is set to 0.005 and the number of operation steps is set to 200.
(3) And optimizing the parameter setting of the convolutional neural network model according to the performance of the convolutional neural network model on the verification set under different parameter combinations, so that the performance of the convolutional neural network model on the verification set is continuously improved, the optimal convolutional neural network model weight is obtained, and the moisture content parameter measurement is completed.
The application of the water content measuring method based on deep learning in oil well exploitation comprises the following steps:
(1) starting a four-sector conductivity sensor, measuring fluid data of oil-water two-phase flow in a wellhead oil-gas pipeline, uploading the fluid data to an upper computer, and obtaining a sample at the current moment according to the water content measuring method based on deep learning;
(2) taking the current time sample as the input of a convolutional neural network model, and obtaining the water content output of the current sample through the calculation and analysis of the convolutional neural network model;
(3) the water content output is associated with the corresponding sample and stored in an upper computer as the current processing information;
(4) and (4) repeating the steps (2) to (3), continuously obtaining the water content value corresponding to the fluid sample in real time, storing the water content value as the state information of the current wellhead oil and gas pipeline until the four-sector conductivity sensor is closed, and analyzing historical data to obtain the total water content in the measuring time.
The above description of the present invention and the embodiments is not limited thereto, and the description of the embodiments is only one of the implementation manners of the present invention, and any structure or embodiment similar to the technical solution without inventive design is within the protection scope of the present invention without departing from the inventive spirit of the present invention.
Claims (5)
1. A moisture content measuring method based on deep learning is characterized by comprising the following steps:
1) collecting fluid data of different working conditions in a wellhead oil and gas pipeline by using a four-sector conductivity sensor, acquiring a label, and uploading the label to an upper computer for storage;
2) constructing a data set, specifically, after preprocessing fluid data, obtaining a sample from the fluid data by using a sliding window with overlap, adding a corresponding label to the sample, taking 80% of the sample as a training set, 10% of the sample as a verification set, and taking the remaining 10% of the sample as a test set;
3) constructing a convolutional neural network model for completing the feature extraction and the water content value prediction of the fluid sample; the convolutional neural network model is used for extracting spatial information and time domain characteristics in two-phase flow data and measuring the water content through the multitask output module; the convolutional neural network model has 19 layers and comprises 4 convolutional modules and 1 multitask output module; the layers 1 to 4 are first convolution modules, the layers 5 to 8 are second convolution modules, the layers 9 to 12 are third convolution modules, the layers 13 to 15 are fourth convolution modules, and the layers 16 to 19 are multitask output modules; wherein,
the first convolution module, the second convolution module and the third convolution module have the same structure and respectively comprise: a temporal convolution layer, a spatial convolution layer, an average pooling layer, and a discard layer; the fourth convolution module comprises the following components which are connected in sequence: a temporal convolution layer, a spatial convolution layer, and a discard layer; wherein,
(1) the time convolution layer is used for filtering the four-channel samples along the time dimension and extracting time domain characteristics; the time convolution layer uses
Each convolution kernel is corresponding to a first convolution module, a second convolution module, a third convolution module and a fourth convolution module, wherein k is 1,2,3 and 4, and different feature maps are obtained after each convolution kernel and the input sample are subjected to convolution operation to obtain the different feature maps
A feature map; the formula for this layer convolution is as follows:
wherein,
is an input for the time convolution layer,
is a characteristic map of the time convolution layer output, fTIs the activation function of the time convolution layer, kTIndex of points in the convolution kernel for time convolution layer, GTFeature dimension, g, input for time convolution layerTThe feature index input for the time convolution layer,
and
respectively, the p-th in the time convolution layerTA convolution kernel and an offset, wherein the offset is calculated by the convolution kernel,
is the size of the time convolution layer convolution kernel, where k is 1,2,3,4, corresponding to the first, second, third and fourth convolution modules, respectively, StTStep size of time convolution layer convolution kernel, subTSubscript shorthand for time convolution layer input, mT,nTA location index of features extracted for the time convolution layer;
the boundary value is extended by complementing 0, and the size of each feature graph is IT/StT×STIn which IT,STIs the dimension of the time convolution layer input; for the feature after convolutionCarrying out batch normalization processing to accelerate convergence;
(2) the space convolution layer is used for integrating the samples of the four channels along the space dimension to obtain space information; the space convolution layer is used
A convolution kernel, wherein k is 1,2,3,4, respectively corresponding to the first convolution module, the second convolution module, the third convolution module and the fourth convolution module,
the convolution cores integrate the input feature maps to obtain a total
The characteristic diagram has the following formula:
wherein,
is an input for the space convolution layer,
is a characteristic diagram of the spatial convolution layer output, fCIs the activation function of the space convolution layer, kCIndex of points in convolution kernel for spatial convolution layer, GCFeature dimension, g, input for space convolution layerCThe feature index input for the spatial convolution layer,
and
respectively being the p-th in the space convolution layerCA convolution kernel and an offset, LeCAnd StCRespectively, the size and step size of the convolution kernel of the space convolution layer, subCSubscript abbreviation, m, of space convolution layer inputC,nCA location index of features extracted for the spatial convolution layer;
the input boundary value is extended by complementing 0, and the size of each feature map is IC/StC×SCIn which IC,SCThe dimension of the input dimension of the space convolution layer is used for carrying out batch normalization processing on the feature graph after convolution so as to accelerate convergence;
(3) the average pooling layer is used for selecting representative values in the pooling area to reduce characteristic dimensionality, removing redundant information and simplifying subsequent calculation, and a formula is described as follows:
wherein,
to average out the characteristics of the pooled layer output,
in order to average the inputs to the pooling layer,
for averaging the size of the pooling kernel, where k is 1,2,3, corresponding to the first, second and third convolution modules, gA,mA,nAIndex for average pooling layer characteristics, kAAn index of points in the average pooled kernel;
(4) a discarding layer, wherein a part of neurons and connection thereof are discarded randomly according to a discarding rate, and a dense neural network model is forced to learn more global characteristics;
the multitask output module comprises a flattening layer, a discarding layer and a full connection layer which are sequentially connected, wherein the output of the full connection layer is respectively connected with a numerical value full connection layer and an auxiliary full connection layer; wherein:
(1) the flattening layer is used for integrating input multi-dimensional features into one-dimensional comprehensive information; the output dimension of the layer is IF×SF×NFIn which IF,SFIs the dimension of the size of the input of the flattening layer, NFIs the number dimension of the flattening layer input;
(2) the discarding layer discards part of neurons and their connections randomly according to a discarding rate, and has an output dimension of
Is provided with
(3) The full connection layer obtains brand new output characteristics by weighting and recombining the input characteristics; the formula is as follows:
wherein,
for the features of the full link layer output,
for input of full connection layer, fDFor the activation function of the fully connected layer,
the number of neurons input to the fully connected layer,
weight and offset, k, of the full connection layer, respectivelyDInput index, g, for the full connection layerDThe index is output for the full-link layer,
is the number of output neurons;
(4) the numerical fully-connected layer is provided with 1 neuron, and the activation function of the layer is Sigmoid; the auxiliary full connection layer has Nu19A neuron, the layer activation function being Softmax;
4) and optimizing, training and adjusting parameters of the convolutional neural network model to obtain the optimal convolutional neural network model under the current architecture, and completing parameter measurement.
2. The deep learning-based water cut measuring method according to claim 1, wherein the step 1) comprises:
(1) connecting the four-sector conductivity sensor arranged on the transparent pipe section with a wellhead oil and gas acquisition pipeline through a flange, acquiring fluid data of oil-water two-phase flow under different working conditions in the wellhead oil and gas pipeline, and uploading the fluid data to an upper computer for storage;
(2) collecting an oil-water sample flowing through a transparent pipe section in the working time of the four-sector conductivity sensor to obtain a water content label of the oil-water sample, determining the overall range of the water content according to the characteristics of an oil well, taking 3% of the water content in the overall range as a step point, and selecting fluid data for subsequent model training.
3. The method for measuring the water cut based on deep learning according to claim 1, wherein the step 2) comprises:
(1) fluid data is pre-processed, with the following formula:
wherein,
is the ith fluid data of the jth sector,
and
respectively the mean and standard deviation of the j-th sector fluid data,
the ith fluid data of the preprocessed jth sector;
(2) intercepting fluid data of each working condition as a single sample through a sliding window with overlap, wherein the label of the sample refers to a water content label, each group of water content labels is regarded as one type, obtaining a class label of the sample, and setting the length of the sliding window as M and the overlap rate as PoThen the single sample is a two-dimensional matrix of M × 4, and is obtained from the fluid data with length L
A sample of, wherein,
the representative data is rounded down, and N samples of the fluid data under all working conditions are obtained;
(3) randomly using 80% of N samples as training set, 10% as verification set and the rest 10% as test set.
4. The method for measuring the water cut based on the deep learning as claimed in claim 1, wherein the step 4) comprises:
(1) constructing a loss function between the label and the output of the convolutional neural network model, describing the difference between the label and the output by using the loss function, and guiding the convolution according to the principle of minimizing the differenceOptimizing a neural network model; loss L of convolutional neural network modeltotalComprises two parts, respectively: a) loss of numerically fully connected layers LpredSelecting the average absolute error as a loss function; b) loss L of auxiliary full link layerclaSelecting a cross entropy loss function; loss LtotalThe expression is as follows:
Lcla＝-∑p(x)×log q(x)
Ltotal＝α×Lpred+β×Lcla
wherein, yi,
Respectively labeling and outputting the water content of the convolutional neural network model, wherein n is the number of samples; p (x), q (x) are the probability distributions of the class label and class output, respectively; α, β are the weights of the two-part loss function;
(2) setting parameters of each layer in the convolutional neural network model, performing iterative training on the weight of the convolutional neural network model under the parameter combination by adopting an Adam-based gradient descent method, and updating the weight of the network by a gradient back propagation method; during training, the learning rate is set to LrThe number of running steps is set to nepoch；
(3) And optimizing the parameter setting of the convolutional neural network model according to the performance of the convolutional neural network model on the verification set under different parameter combinations, so that the performance of the convolutional neural network model on the verification set is continuously improved, the optimal convolutional neural network model weight is obtained, and the moisture content parameter measurement is completed.
5. The application of the deep learning based water cut measuring method in oil well exploitation is characterized by comprising the following steps:
(1) starting a four-sector conductivity sensor, measuring fluid data of oil-water two-phase flow in a wellhead oil-gas pipeline, uploading the fluid data to an upper computer, and obtaining a sample at the current moment according to the water content measuring method based on deep learning;
(2) taking the current time sample as the input of a convolutional neural network model, and obtaining the water content output of the current sample through the calculation and analysis of the convolutional neural network model;
(3) the water content output is associated with the corresponding sample and stored in an upper computer as the current processing information;
(4) and (4) repeating the steps (2) to (3), continuously obtaining the water content value corresponding to the fluid sample in real time, storing the water content value as the state information of the current wellhead oil and gas pipeline until the four-sector conductivity sensor is closed, and analyzing historical data to obtain the total water content in the measuring time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010481478.6A CN111638249B (en) | 2020-05-31 | 2020-05-31 | Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010481478.6A CN111638249B (en) | 2020-05-31 | 2020-05-31 | Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111638249A CN111638249A (en) | 2020-09-08 |
| CN111638249B true CN111638249B (en) | 2022-05-17 |
Family
ID=72332165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010481478.6A Active CN111638249B (en) | 2020-05-31 | 2020-05-31 | Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111638249B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112485304B (en) * | 2020-12-14 | 2022-08-09 | 河北工业大学 | Milk detection method and system based on conductance method |
| CN112814649B (en) * | 2021-01-04 | 2023-03-10 | 天津科技大学 | Intelligent metering device and method for water content of oil well produced liquid |
| CN112700051A (en) * | 2021-01-04 | 2021-04-23 | 天津科技大学 | Res-TCN neural network-based intelligent prediction method for oil well liquid production associated gas |
| CN112733449B (en) * | 2021-01-11 | 2022-12-02 | 中国海洋大学 | CNN well-seismic joint inversion method, CNN well-seismic joint inversion system, CNN well-seismic joint inversion storage medium, CNN well-seismic joint inversion equipment and CNN well-seismic joint inversion application |
| CN115060744A (en) * | 2022-04-29 | 2022-09-16 | 天津大学 | Two-phase flow parameter measurement method based on space-time parallel feature fusion network |
| CN115032244A (en) * | 2022-04-29 | 2022-09-09 | 天津大学 | Multi-electrode two-phase flow parameter measuring method based on time channel fusion network |
| CN115031794B (en) * | 2022-04-29 | 2024-07-26 | 天津大学 | Novel gas-solid two-phase flow measuring method based on multi-feature graph convolution |
| CN114878599B (en) * | 2022-04-29 | 2024-10-18 | 天津大学 | Oil-water two-phase flow water content measuring method based on time-frequency chart convolution |
| CN115902004B (en) * | 2022-11-08 | 2023-08-18 | 中科特肯(山东)智能科技有限公司 | Measurement device and measurement method for conductivity of degassed hydrogen |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650929A (en) * | 2016-10-11 | 2017-05-10 | 天津大学 | Recursive-graph-based deep learning model and its application in oil-water phase rate measurement |
| CN110069958A (en) * | 2018-01-22 | 2019-07-30 | 北京航空航天大学 | A kind of EEG signals method for quickly identifying of dense depth convolutional neural networks |
| CN110222643A (en) * | 2019-06-06 | 2019-09-10 | 西安交通大学 | A kind of Steady State Visual Evoked Potential Modulation recognition method based on convolutional neural networks |
| CN110632132A (en) * | 2019-07-09 | 2019-12-31 | 东营智图数据科技有限公司 | High-yield gas-oil well wellhead liquid water content prediction method based on multi-sensor measurement and deep convolutional neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10224058B2 (en) * | 2016-09-07 | 2019-03-05 | Google Llc | Enhanced multi-channel acoustic models |
-
2020
- 2020-05-31 CN CN202010481478.6A patent/CN111638249B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650929A (en) * | 2016-10-11 | 2017-05-10 | 天津大学 | Recursive-graph-based deep learning model and its application in oil-water phase rate measurement |
| CN110069958A (en) * | 2018-01-22 | 2019-07-30 | 北京航空航天大学 | A kind of EEG signals method for quickly identifying of dense depth convolutional neural networks |
| CN110222643A (en) * | 2019-06-06 | 2019-09-10 | 西安交通大学 | A kind of Steady State Visual Evoked Potential Modulation recognition method based on convolutional neural networks |
| CN110632132A (en) * | 2019-07-09 | 2019-12-31 | 东营智图数据科技有限公司 | High-yield gas-oil well wellhead liquid water content prediction method based on multi-sensor measurement and deep convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111638249A (en) | 2020-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111638249B (en) | Water content measuring method based on deep learning and application of water content measuring method in oil well exploitation | |
| CN110533631B (en) | SAR image change detection method based on pyramid pooling twin network | |
| CN111649779B (en) | Oil well oil content and total flow rate measuring method based on dense neural network and application | |
| CN112001270B (en) | Ground radar automatic target classification and identification method based on one-dimensional convolutional neural network | |
| CN108399248A (en) | A kind of time series data prediction technique, device and equipment | |
| CN115758212B (en) | Mechanical equipment fault diagnosis method based on parallel network and transfer learning | |
| CN109165660A (en) | A kind of obvious object detection method based on convolutional neural networks | |
| CN111832227B (en) | Shale gas saturation determination method, device and equipment based on deep learning | |
| CN113095409A (en) | Hyperspectral image classification method based on attention mechanism and weight sharing | |
| CN114239397A (en) | Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning | |
| CN112147432A (en) | BiLSTM module based on attention mechanism, transformer state diagnosis method and system | |
| CN112949944B (en) | Intelligent groundwater level prediction method and system based on space-time characteristics | |
| Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
| CN114492184A (en) | Method and device for predicting remaining service life of aero-engine through time-frequency domain analysis | |
| CN116343046A (en) | Multi-source remote sensing urban river water quality monitoring method based on self-optimizing machine learning | |
| CN113592939A (en) | Deep learning method for judging size of narrow blood vessel based on coronary angiography image | |
| CN113756786B (en) | Oil pumping well time sequence indicator diagram prediction method based on deep learning | |
| CN110260914B (en) | Engineering safety monitoring system region division method based on measuring point space-time characteristics | |
| CN112885378B (en) | Speech emotion recognition method and device and storage medium | |
| CN117470877A (en) | Oil well oil-water two-phase flow parameter measurement method based on array spiral microwave sensor | |
| Ju et al. | Hydrologic simulations with artificial neural networks | |
| CN115031794B (en) | Novel gas-solid two-phase flow measuring method based on multi-feature graph convolution | |
| CN115963788A (en) | Multi-sampling-rate industrial process key quality index online prediction method | |
| Xu et al. | Prediction of the Wastewater's pH Based on Deep Learning Incorporating Sliding Windows. | |
| CN115909086A (en) | SAR target detection and identification method based on multistage enhanced network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |