CN109209361B - Method for predicting stratum parameters of fractured ultra-low permeability reservoir - Google Patents
Method for predicting stratum parameters of fractured ultra-low permeability reservoir Download PDFInfo
- Publication number
- CN109209361B CN109209361B CN201810996052.7A CN201810996052A CN109209361B CN 109209361 B CN109209361 B CN 109209361B CN 201810996052 A CN201810996052 A CN 201810996052A CN 109209361 B CN109209361 B CN 109209361B
- Authority
- CN
- China
- Prior art keywords
- tracer
- output
- time domain
- layer neuron
- tracing
- 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
- 230000035699 permeability Effects 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 29
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 139
- 239000000700 radioactive tracer Substances 0.000 claims abstract description 121
- 238000004519 manufacturing process Methods 0.000 claims abstract description 111
- 238000009826 distribution Methods 0.000 claims abstract description 49
- 238000013528 artificial neural network Methods 0.000 claims abstract description 45
- 239000000126 substance Substances 0.000 claims abstract description 39
- 238000013178 mathematical model Methods 0.000 claims abstract description 26
- 238000005516 engineering process Methods 0.000 claims abstract description 22
- 238000003062 neural network model Methods 0.000 claims abstract description 22
- 238000012544 monitoring process Methods 0.000 claims abstract description 16
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 8
- 210000002569 neuron Anatomy 0.000 claims description 159
- 230000003068 static effect Effects 0.000 claims description 40
- 238000002347 injection Methods 0.000 claims description 34
- 239000007924 injection Substances 0.000 claims description 34
- 239000013598 vector Substances 0.000 claims description 24
- 238000004146 energy storage Methods 0.000 claims description 14
- 239000006185 dispersion Substances 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000005465 channeling Effects 0.000 claims description 4
- 239000003129 oil well Substances 0.000 claims description 4
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000013146 percutaneous coronary intervention Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000002452 interceptive effect Effects 0.000 claims description 2
- 238000011161 development Methods 0.000 abstract description 8
- 238000012360 testing method Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 239000003795 chemical substances by application Substances 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mining & Mineral Resources (AREA)
- Geology (AREA)
- Fluid Mechanics (AREA)
- Environmental & Geological Engineering (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geochemistry & Mineralogy (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明涉及裂缝性特低渗透油藏开发技术领域，尤其涉及一种裂缝性特低渗透油藏地层参数预测方法，首先建立时域化学示踪物理模型；其次建立时域化学示踪数学模型；再次建立时域示踪神经网络模型；最后根据时域化学示踪物理模型、时域化学示踪数学模型和时域示踪神经网络模型，利用BP神经网络技术求解时域化学示踪数学模型，根据求解结果预测裂缝性低渗透油藏的裂缝条带个数N、裂缝条带平均渗透率K、注入水波及体积V、注入水分配系数f和水驱推进速度v。本发明的预测方法能够利用注采井组历次示踪监测解释结果来预测未来井间地层参数，实现裂缝性特低渗油藏参数在时间上的推移，反馈参数的变化趋势和规律。
The invention relates to the technical field of development of fractured ultra-low permeability oil reservoirs, in particular to a method for predicting formation parameters of fractured ultra-low permeability oil reservoirs. The time-domain tracer neural network model is established again; finally, according to the time-domain chemical tracer physical model, the time-domain chemical tracer mathematical model and the time-domain tracer neural network model, the BP neural network technology is used to solve the time-domain chemical tracer mathematical model, According to the solution results, the number of fractured low-permeability reservoirs N, the average permeability of fractured strips K, the swept volume of injected water V, the distribution coefficient of injected water f and the waterflooding propulsion speed v are predicted. The prediction method of the invention can predict the future inter-well formation parameters by using the tracer monitoring and interpretation results of the injection-production well group, realize the time passage of the fractured ultra-low permeability oil reservoir parameters, and feedback the changing trend and law of the parameters.
Description
Technical Field
The invention relates to the technical field of fractured ultra-low permeability reservoir development, in particular to a method for predicting stratum parameters of a fractured ultra-low permeability reservoir.
Background
In the water drive development process of the fractured ultra-low permeability reservoir, the formation parameters between injection wells and production wells can generate obvious time-domain changes, and the dynamic change rule and trend of the formation parameters are difficult to effectively characterize and predict. Along with the continuous deepening of the water injection development of the fractured ultra-low-permeability oil reservoir, the stratum parameters can generate time-domain change, and particularly the water drive development effect of the whole injection and production well group is changed due to the change of the water injection strength, the injection pressure and the like of the fractures in the stratum. How to predict future interwell stratum parameters by using the historical tracing monitoring interpretation results of injection and production well groups to realize the time lapse of fractured ultra-low permeability reservoir parameters and the feedback of the variation trend and the law of the parameters becomes the key for realizing the dynamic real-time monitoring and prediction of the fractured ultra-low permeability reservoir water injection development.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a fractured ultra-low permeability reservoir stratum parameter prediction method, which can predict future interwell stratum parameters by using the historical tracing, monitoring and interpretation results of injection and production well groups, and realize the time lapse of fractured ultra-low permeability reservoir parameters and the change trend and rule of feedback parameters.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a method for predicting stratum parameters of a fractured ultra-low permeability reservoir comprises the following steps:
step 1, establishing a time domain chemical tracing physical model, utilizing well group historical tracer explanation parameters, production well and oil production well production dynamic parameters to combine with related static parameters to realize the transition and prediction of the tracer explanation parameters in time, selecting known dynamic historical information and known static information which visually indicate fracture ultra-low permeability reservoir formation parameters, and definitely solving variables;
step 2, establishing a time domain chemical tracing mathematical model, taking the time domain chemical tracing physical model established in the step 1 and history information monitored by the tracer as a learning sample, searching the inherent complex implicit relation among all parameters, and establishing mathematical expression describing the physical model;
step 3, establishing a time domain tracing neural network model, performing time domain tracing analysis by using a BP neural network, and establishing a more appropriate network structure, wherein the network structure comprises an input layer node number, a hidden layer node number and an output layer node number;
and 4, solving the time domain chemical tracing mathematical model by using a BP neural network technology according to the time domain chemical tracing physical model, the time domain chemical tracing mathematical model and the time domain tracing neural network model, and predicting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the fractured low-permeability oil reservoir according to the solving result.
The time domain chemical tracing physical model and the time domain chemical tracing mathematical model both belong to a time domain tracing theoretical model, and the time domain tracing is to analyze and predict the change trend of future tracer production curve interpretation parameters according to the parameter values interpreted by tracing monitoring historical curves and by combining production dynamic data and geological static data, so as to guide the adjustment of the oilfield flooding development scheme. The chemical tracing monitoring historical data between wells is taken as a data source and is regarded as a time sequence, a physical model and a mathematical model which accord with the actual condition of an oil field are firstly established by utilizing the neural network technology to carry out time-lapse tracing prediction and analysis, and simultaneously, the definite solution condition is determined according to the actual production.
The time domain chemical tracing physical model in the step 1 comprises known dynamic history information, known static information, solving variables and a prediction technology, wherein:
known dynamic history information includes: the production time t, the yield, the water content, the well group accumulated injection-production ratio, the well group stage injection-production ratio, the number of the last time of fracture strips, the average permeability of the last time of fracture strips, the last time of water injection swept volume, the last time of water injection distribution coefficient and the last time of water drive propulsion speed;
known static information includes: the static productivity coefficient of the bedrock at the production section, the static energy storage coefficient of the bedrock at the production section and the injection-production well spacing;
solving the variables includes: the number of the crack strips, the average permeability of the crack strips, the swept volume of injected water, the distribution coefficient of the injected water and the water drive propelling speed are controlled;
the prediction technology is a neural network technology.
In the known dynamic historical information, the number of previous fracture strips, the average permeability of the previous fracture strips, the water wave and volume of the previous injection, the distribution coefficient of the previous injection water and the previous water drive propulsion speed are calculated by adopting a fractured ultra-low permeability reservoir tracer classified interpretation model, and the fractured ultra-low permeability reservoir tracer classified interpretation model comprises a single-peak tracer interpreted physical model, a multi-peak tracer interpreted physical model or a wide-platform tracer interpreted physical model.
The single-peak tracer explained physical model is equivalent to an artificial crack channeling model, in the single-peak tracer explained physical model, 1 crack strip is a flow tube bundle consisting of n flow tubes with the length of L and the equivalent diameter of D, tracer flows in the n flow tubes in the crack strip, and the tracer explained model is as follows:
in the formula: c is the output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n is the total number of equivalent flow tubes; vdThe total injected volume of the tracer slug; alpha is hydrodynamic dispersivity; l is the equivalent length of the flow tube; d is the equivalent diameter of the equivalent flow tube; q is the average daily injection;
the multimodal tracer explained physical model is a differential fracture interactive model, a plurality of fracture strips are arranged in the multimodal tracer explained physical model, and the ith fracture strip is distributed from niEach length is LiAn equivalent diameter of DiThe tracer is n in the plurality of fracture stripsiFlow in individual flowtubes, tracer interpretation model is:
in the formula: c is the output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n isiThe number of equivalent flow tubes of the ith crack strip; vdThe total injected volume of the tracer slug; q is the average daily injection(ii) a Di is the equivalent diameter of any equivalent flow pipe of the ith fracture strip; li is the equivalent flow tube length of the ith crack strip; alpha is alphaiThe hydrodynamic dispersion constant of the tracer in the ith fracture strip equivalent flow tube is obtained;
the physical model of the wide-platform tracer explanation is a relatively uniform crack propulsion type, and is equivalent to a plurality of crack strips, and the ith crack strip is distributed from niEach length is LiAn equivalent diameter of DiThe flow tube bundle formed by the flow tubes, the tracer arrives at the oil well in different flow tubes in sequence, and the tracer interpretation model is as follows:
in the formula: c is the output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n isiThe number of equivalent flow tubes of the ith crack strip; vdThe total injected volume of the tracer slug; q is the average daily injection; di is the equivalent diameter of any equivalent flow pipe of the ith fracture strip; li is the equivalent flow tube length of the ith crack strip; alpha is alphaiThe hydrodynamic dispersion constant of the tracer in the ith fracture strip equivalent flow tube is obtained; m is the equivalent flow resistance constant.
In the known static information, the static productivity coefficient PC of the bedrock at the production section is equal to the product of the effective permeability k of the bedrock at the production section and the effective thickness h of the bedrock at the production section, and the static energy storage coefficient NH of the bedrock at the production section is equal to the effective thickness h of the bedrock at the production section and the effective porosity of the bedrock at the production section
And oil saturation S of production zone bedrockoThe product of (a).
When a time domain tracing mathematical model is established, regarding a tracing monitoring process as a time point, and taking historical information monitored by a tracer as a learning sample, wherein the time domain chemical tracing mathematical model is as follows:
in the formula, z is a time domain tracing BP neural network learning sample serial number; n is the number of crack strips; k is the average permeability of the crack band; v is the volume of injected water wave; f is the distribution coefficient of injected water; v is the water drive propulsion speed; q is the yield; f. ofwThe water content is obtained; r is the injection-production ratio of the well group stage; RC is the cumulative injection-production ratio of the well group; w is the weight and the connection coefficient between two adjacent layers.
The time domain tracing neural network model is of a three-layer network structure and comprises an input layer, a hidden layer and an output layer group, wherein the hidden layer is of a 1-layer structure; the output layer contains 5 nodes, and 5 nodes are respectively: the number N of the crack strips, the average permeability K of the crack strips, the wave spread volume V of injected water, the distribution coefficient f of the injected water and the water drive propulsion speed V.
The input layer contains 8 parameters corresponding to the prediction time, and the 8 parameters are respectively: yield q, water content fwThe method comprises the following steps of (1) well group accumulated injection-production ratio RC, well group stage injection-production ratio R, production time t, injection-production well spacing L, production section bedrock static productivity coefficient PC and production section bedrock static energy storage coefficient NH;
or the input layer contains the following parameters: yield q, water content fwThe method comprises the following steps of well group accumulated injection-production ratio RC, well group stage injection-production ratio R, production time t, injection-production well spacing L, production section bedrock static productivity coefficient PC, production section bedrock static energy storage coefficient NH, last fracture strip number N, last fracture strip average permeability K, last injected water wave volume V, last injected water distribution coefficient f and last water drive propulsion speed V.
In the step 4, the process of solving the time domain chemical tracing mathematical model by using the BP neural network technology comprises the following steps:
step 4.1, carrying out normalization processing on the time domain tracing BP neural network learning sample, and setting all weights of the time domain chemical tracing mathematical model as smaller random numbers;
step 4.2, given an input vector X and a desired output vector d: sending the X to an input layer of the time domain tracing neural network model;
4.3, calculating an output vector Y of a hidden layer of the time domain tracing neural network model;
4.4, calculating an output vector O of an output layer of the time domain tracing neural network model;
step 4.5, calculating an error signal delta of an output layer of the time domain tracing neural network modelo；
Step 4.6, calculating an error signal delta of a hidden layer of the time domain tracing neural network modely；
Step 4.7, error signal delta of output layer is utilizedoAnd error signal delta of the hidden layeryAdjusting and correcting the weight;
step 4.8, calculating the error of the BP neural network system;
step 4.9, judging whether the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency;
if the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency, outputting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propulsion speed V of the low-permeability reservoir;
and if the error of the BP neural network system does not meet the condition of the preset precision or exceeds the maximum learning frequency, repeating the step 4.3 to the step 4.9 until the error of the BP neural network system meets the condition of the preset precision or exceeds the maximum learning frequency, and finally outputting the number N of the fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the low-permeability reservoir.
In step 4.2, the input vector X and the expected output vector d are respectively:
X＝[qI fwI RCI RI TI L PC NH]T
d＝[NI KI VI fI vI]T
wherein q isIYield of learning samples, f, for the ith time domain tracewIIs the I timeWater cut, RC, of domain tracer learning samplesICumulative injection-production ratio, R, for the well group of the I time domain tracer learning samplesIWell group stage injection-production ratio, T, for the I time domain tracer learning sampleITracing the production time of the learning sample for the I time domain, wherein L is the injection-production well spacing, PC is the static productivity coefficient of the bedrock of the production section, NH is the static energy storage coefficient of the bedrock of the production section, and N isINumber of crack bands, K, for the I time domain trace learning sampleIAverage permeability of crack bands, V, for the I time domain trace learning sampleIVolume of injected water wave, f, for the I time domain tracer learning sampleIDistribution coefficient of injected water, v, for the I time domain tracer learning sampleITracing the water drive propulsion speed of the learning sample for the I time domain;
in step 4.3, the output vector Y is:
wherein A is the threshold value of each neuron of the hidden layer, viqThe weight value between the input layer neuron i and the hidden layer neuron q is obtained; v. offwkFor input layer neuron i and hidden layer neuron fwThe weight value between; v. ofRCkThe weight value between the input layer neuron i and the hidden layer neuron RC is obtained; v. ofRkIs the weight between input layer neuron i and hidden layer neuron R; v. ofTkThe weight value between the input layer neuron i and the hidden layer neuron T is obtained; v. ofLkIs the weight value between the input layer neuron i and the hidden layer neuron L; v. ofPCkThe weight value between the input layer neuron i and the hidden layer neuron PC is obtained; v. ofNHkThe weight value between the input layer neuron i and the hidden layer neuron NH is obtained;
in step 4.4, the output vector O is:
wherein B is the threshold of each neuron of the output layer, wjNThe weight value between the output layer neuron N and the hidden layer neuron j is obtained; w is ajKThe weight value between the output layer neuron K and the hidden layer neuron j is obtained; w is ajVThe weight value between the output layer neuron V and the hidden layer neuron j is obtained; w is ajfThe weight value between the output layer neuron f and the hidden layer neuron j is obtained; w is ajvThe weight value between the output layer neuron v and the hidden layer neuron j is obtained; y isjThe jth output value of the hidden layer;
in step 4.5, the error signal delta of the layer is outputoComprises the following steps:
δo＝[oN(NI-oN)(1-oN)oK(KI-oK)(1-oK)
oV(VI-oV)(1-oV)of(fI-of)(1-of)
ov(vI-ov)(1-ov)]T
wherein o isNThe output value is the number of the crack strips of the output layer; oKThe output value is the average permeability of the crack strip of the output layer; oVInjecting water waves and volume output values into the output layer; ofInjecting an output value of the water distribution coefficient for the output layer; ovThe output value is the water drive propulsion speed of the output layer;
in step 4.6, the error signal δ of the hidden layeryComprises the following steps:
wherein, deltak oIs the error signal of the output layer; w is aqkThe weight value between the output layer neuron k and the hidden layer neuron q is obtained; w is afwkFor output layer neuron k and hidden layer neuron fwThe weight value between; w is aRCkThe weight value between the output layer neuron k and the hidden layer neuron RC is obtained; w is aRkIs the weight between the output layer neuron k and the hidden layer neuron R; w is aTkAs output layer neuron k and hidden layer neuronThe weight between T; w is aNHkThe weight value between the output layer neuron k and the hidden layer neuron NH is obtained; w is aPCkThe weight value between the output layer neuron k and the hidden layer neuron PC is obtained; w is aLkThe weight value between the output layer neuron k and the hidden layer neuron L is obtained; y isqIAn output value for the implicit layer neuron qth sample; y isfwIFor hidden layer neurons fwThe output value of the ith sample; y isRCIAn output value for the implicit layer neuron RC sample I; y isRIAn output value for the I < th > sample of hidden layer neuron R; y isTIAn output value for the hidden layer neuron Tth sample; y isPCIs the output value of the hidden layer neuron PC; y isLIs the output value of hidden layer neuron L;
in step 4.7, the error signal delta of the output layer is usedoAnd error signal delta of the hidden layeryAnd (3) adjusting and correcting the weight by adopting the following formula:
wherein, output layer neuron k is 1, 2, …, l; w is ajkThe weight value between the output layer neuron k and the hidden layer neuron j is obtained; output layer neuron j ═ 1, 2, …, m; v. ofijThe weight value between the input layer neuron i and the hidden layer neuron j is obtained; Δ wjkAnd Δ vijAre respectively the weight value wjkAnd vijChanging the value along the negative gradient direction of the output error E; s is the training times; eta is the learning rate; mc is the momentum factor;
in step 4.8, the error of the BP neural network system is:
wherein d isk(I) Outputting the expected output value of the layer neuron k for the I time domain tracing learning sample; ok(I) And outputting the output value of the layer neuron k for the I time domain tracing learning sample.
The learning rate eta ranges from 0.01 to 0.8; the momentum factor mc ranges from 0.85 to 0.95.
Compared with the prior art, the invention has the following beneficial effects:
the method for predicting the stratum parameters of the fractured ultra-low permeability reservoir firstly establishes a time domain chemical tracing physical model; then establishing a time domain chemical tracing mathematical model; then establishing a time domain tracing neural network model; and finally, solving the time domain chemical tracing mathematical model by using a BP neural network technology according to the time domain chemical tracing physical model, the time domain chemical tracing mathematical model and the time domain tracing neural network model, and predicting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the fractured low-permeability oil reservoir according to the solving result. The prediction method can predict the formation parameters between wells in the future by using the historical tracing, monitoring and explaining results of the injection-production well group, and realizes the time lapse of the fractured ultra-low permeability reservoir parameters and the feedback of the change trend and the law of the parameters. Compared with other oil reservoir monitoring technologies, the method can realize time domain of independent oil reservoir monitoring points, can predict interwell stratum parameters at any time, and provides a basis for subsequent development scheme adjustment and water channeling water flooding comprehensive treatment of the fractured ultra-low permeability oil reservoir.
Drawings
FIG. 1 is an explanatory physical model of the unimodal tracer employed in the present invention;
FIG. 2 is a graph illustrating the physical model for a unimodal tracer employed in the present invention;
FIG. 3 illustrates a physical model for a multimodal tracer employed in the present invention;
FIG. 4 is a graph illustrating the physical model for a multimodal tracer employed in the present invention;
FIG. 5 is a broad-table tracer interpretation physical model employed in the present invention;
FIG. 6 is a graph illustrating the physical model for a broadstand tracer used in the present invention;
FIG. 7 is a diagram of a time-domain tracing neural network in embodiment 1 of the present invention;
FIG. 8 is a diagram of a time-domain tracing neural network in embodiment 2 of the present invention;
FIG. 9 is a flow chart of solving the time domain tracing BP neural network adopted by the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Examples
In the embodiment, a time domain tracer analysis simulation is performed by taking an Ordos basin GGY oil field A water injection block 1291-3 injection-production well group as an example.
The method for predicting the stratum parameters of the fractured ultra-low permeability reservoir comprises the following steps:
step 1, establishing a time domain chemical tracing physical model, wherein the time domain chemical tracing physical model mainly comprises 4 parts of known dynamic historical information, known static information, solving variables and a prediction technology, and the method comprises the following specific steps:
(1) known dynamic history information includes: the production time t, the yield, the water content, the well group accumulated injection-production ratio, the well group stage injection-production ratio, the number of the last fracture strips (, the average permeability of the last fracture strips, the water swept volume of the last injection, the water distribution coefficient of the last injection and the water drive propulsion speed of the last time;
(2) known static information includes: the static productivity coefficient of the bedrock at the production section, the static energy storage coefficient of the bedrock at the production section and the injection-production well spacing;
(3) solving the variables includes: the number of the crack strips, the average permeability of the crack strips, the swept volume of injected water, the distribution coefficient of the injected water and the water drive propelling speed are controlled;
(4) the prediction technology is a neural network technology.
In the time domain chemical tracing physical model, known dynamic historical information is as follows: the method is characterized in that 5 parameters of the number of the last fracture strips (, the average permeability of the last fracture strips, the water wave and volume injected last time, the water distribution coefficient injected last time and the water drive propulsion speed injected last time) are calculated by adopting a classification and interpretation model of the fractured ultra-low permeability reservoir tracer, namely a single-peak tracer interpretation physical model, a multi-peak tracer interpretation physical model or a wide-platform tracer interpretation physical model:
firstly, as shown in fig. 1, the unimodal tracer explained physical model is equivalent to an artificial fracture channeling type, that is, 1 fracture strip is a flow tube bundle composed of n flow tubes with length L and equivalent diameter D, the tracer can be regarded as flowing in the n flow tubes in the fracture strip, the curve form of the unimodal tracer explained physical model is shown in fig. 2, and the unimodal tracer explained model is:
in the formula: c is the output concentration of the tracer, and the unit is mg/L; c0Is the initial concentration of tracer in g/cm3(ii) a t is time in d (days); f. ofjThe distribution coefficient of the injected water to the oil production well j can be obtained by the ratio of the output of the tracer agent measured by each oil production well to the total injection amount of the tracer agent; n is the total number of equivalent flow tubes and the unit is one; vdIs the total injection volume of the tracer slug in m3(ii) a Alpha is the hydrodynamic dispersivity, in m; l is the equivalent length of the flow tube, and the unit is m; d is the equivalent diameter of the equivalent flow tube and has the unit of mum; q is the average daily injection in m3/d。
② as shown in figure 3, the physical model of the multi-peak tracer explanation is a differential fracture interaction type, namely a plurality of fracture strips, and the distribution of the ith fracture strip can be regarded as niEach length is LiAn equivalent diameter of DiThe seepage difference between crack strips is large, and the tracer can be regarded as n in the crack stripsiThe curve form of the multimodal tracer explained physical model flowing in the flow pipe is shown in figure 4, and the multimodal tracer explained model is as follows:
in the formula: c is the output concentration of the tracer with the unit ofmg/L；C0Is the initial concentration of tracer in g/cm3(ii) a t is time in units of d; f. ofjThe distribution coefficient of the injected water to the oil production well j can be obtained by the ratio of the output of the tracer agent measured by each oil production well to the total injection amount of the tracer agent; n isiThe number of equivalent flow tubes of the ith crack strip is the unit of the number; vdIs the total injection volume of the tracer slug in m3(ii) a Q is the average daily injection in m3D; di is the equivalent diameter of any equivalent flow pipe of the ith crack strip and has the unit of mu m; li is the equivalent flow tube length of the ith crack strip and is m; alpha is alphaiThe hydrodynamic dispersion constant of the tracer in the ith fracture band equivalent flow tube is given in m.
③ As shown in FIG. 5, the physical model of wide platform tracer interpretation is a relatively uniform crack propulsion type, which is equivalent to a plurality of crack strips, and the distribution of the ith crack strip can be seen from niEach length is LiAn equivalent diameter of DiThe tracer sequentially reaches the oil well in different flow pipes, and the curve form of the wide-platform tracer explaining the physical model is shown in figure 6. The broad-table tracer interpretation model is:
in the formula: c is the output concentration of the tracer, and the unit is mg/L; c0Is the initial concentration of tracer in g/cm3(ii) a t is time in units of d; f. ofjThe distribution coefficient of the injected water to the oil production well j can be obtained by the ratio of the output of the tracer agent measured by each oil production well to the total injection amount of the tracer agent; n isiThe number of equivalent flow tubes of the ith crack strip is the unit of the number; vdIs the total injection volume of the tracer slug in m3(ii) a Q is the average daily injection in m3D; di is the equivalent diameter of any equivalent flow pipe of the ith crack strip and has the unit of mu m; li is the equivalent flow tube length of the ith crack strip and is m; alpha is alphaiEquivalent flow for tracer at ith fracture bandThe hydrodynamic dispersion constant in the tube, in m; m is equivalent flow resistance constant and has a unit of M/mum4。
In the time domain chemical tracing physical model, known static historical information comprises: the static productivity coefficient PC of the bedrock is equal to the product of the effective permeability k and the effective thickness h, the static energy storage coefficient NH of the bedrock is equal to the effective thickness h, and the effective porosity
And oil saturation SoThe product of (a).
In the time domain chemical tracing physical model, solved variable parameters are as follows: the number of crack strips, the average permeability of the crack strips, the wave-spread volume of injected water, the distribution coefficient of the injected water and the propulsion speed of water flooding.
In the time domain chemical tracing physical model, the prediction technology is a neural network technology.
Step 2, establishing a time domain tracing mathematical model:
when a time domain tracing mathematical model is established, because the interwell tracing monitoring is the distribution of injected fluid under the stable pressure of the stratum, the stratum seepage characteristics basically keep unchanged, the normal production of an oil well is hardly influenced by the tracing monitoring, and the tracing monitoring process is regarded as a time point. Historical information monitored by the tracer is used as a learning sample, and an inherent complex implicit relation among all parameters is searched, so that each tracing monitoring is continuous, and a time domain tracing mathematical model is as follows:
in the formula, z is a time domain tracing BP neural network learning sample serial number; n is the number of the crack strips, and the unit is one; k is the average permeability of the fractured band and has a unit of 10-3Mu m; v is the volume of injected water wave and is given in m3(ii) a f is the distribution coefficient of injected water, and the unit is f; v water drive propulsion speed, unit is m/d; q is yield, and the unit is t/d; f. ofwIs the water content, unit is%; r is a well group stage injection-production ratio with the unit of f; RC is well group cumulative injectionSampling ratio with unit f; w is the weight, which is the connection coefficient between two adjacent layers.
Step 3, establishing a time domain tracing neural network model
The network structure of the time domain neural network model comprises three layers: namely an input layer, a hidden layer and an output layer. The hidden layer comprises 1 layer; the output layer sets 5 nodes as: the number N of the crack strips, the average permeability K of the crack strips, the wave spread volume V of injected water, the distribution coefficient f of the injected water and the water drive propulsion speed V. Two schemes are designed for an input layer, and only 8 parameters corresponding to the prediction time are considered in scheme 1: i.e. yield q, water content fwThe method comprises the following steps of (1) well group accumulated injection-production ratio RC, well group stage injection-production ratio R, production time t, injection-production well spacing L, production section bedrock static productivity coefficient PC and production section bedrock static energy storage coefficient NH; as shown in fig. 7, taking scheme 1 as an example, the number n of vectors input by the input layer is 8, the number l of vectors output by the output layer is 5, the number of hidden layers is 1, and the number m of hidden points is 8;
in addition to the 8 parameters corresponding to the predicted time, the scheme 2 also considers the 5 parameters of the last trace monitoring explanation: in other words, the number N of fracture strips, the average permeability K of the fracture strips, the swept volume V of the injected water, the distribution coefficient f of the injected water, and the water drive propulsion speed V are also considered, and in case of scheme 2, the number N of vectors input to the input layer is 13, the number l of vectors output from the output layer is 5, the number of hidden layers is 1, and the number m of hidden points is 13.
Step 4, solving the time domain chemical tracing mathematical model by using the BP neural network technology, referring to FIG. 9, wherein the solving step is as follows (the solution is carried out by taking the scheme 1 as an example in the step 4):
step 4.1, carrying out normalization processing on the time domain tracing BP neural network learning sample by using a maximum and minimum method, and setting all weights in a time domain chemical tracing mathematical model as smaller random numbers;
step 4.2, an input vector X and an expected output vector d are given:
X＝[qI fwI RCI RI TI L PC NH]T
d＝[NI KI VI fI vI]T
sending X to the input layer, qIYield of learning samples, f, for the ith time domain tracewIWater cut, RC, of learning samples for the I time domain tracerICumulative injection-production ratio, R, for the well group of the I time domain tracer learning samplesIWell group stage injection-production ratio, T, for the I time domain tracer learning sampleITracing the production time of the learning sample for the I time domain, wherein L is the injection-production well spacing, PC is the static productivity coefficient of the bedrock of the production section, NH is the static energy storage coefficient of the bedrock of the production section, and N isINumber of crack bands, K, for the I time domain trace learning sampleIAverage permeability of crack bands, V, for the I time domain trace learning sampleIVolume of injected water wave, f, for the I time domain tracer learning sampleIDistribution coefficient of injected water, v, for the I time domain tracer learning sampleITracing the water drive propulsion speed of the learning sample for the I time domain;
step 4.3, calculating an output vector Y of the hidden layer:
in the formula, A is the threshold value of each neuron of the hidden layer; v. ofiqThe weight value between the input layer neuron i and the hidden layer neuron q is obtained; v. offwkIs the weight between the input layer neuron i and the hidden layer neuron fw; v. ofRCkThe weight value between the input layer neuron i and the hidden layer neuron RC is obtained; v. ofRkIs the weight between input layer neuron i and hidden layer neuron R; v. ofTkThe weight value between the input layer neuron i and the hidden layer neuron T is obtained; v. ofLkIs the weight value between the input layer neuron i and the hidden layer neuron L; v. ofPCkThe weight value between the input layer neuron i and the hidden layer neuron PC is obtained; v. ofNHkThe weight value between the input layer neuron i and the hidden layer neuron NH is obtained;
step 4.4, calculating an output vector O of the output layer:
in the formula, B is the threshold value of each neuron of the output layer, and B is the threshold value of each neuron of the output layer; w is ajNThe weight value between the output layer neuron N and the hidden layer neuron j is obtained; w is ajKThe weight value between the output layer neuron K and the hidden layer neuron j is obtained; w is ajVThe weight value between the output layer neuron V and the hidden layer neuron j is obtained; w is ajfThe weight value between the output layer neuron f and the hidden layer neuron j is obtained; w is ajvThe weight value between the output layer neuron v and the hidden layer neuron j is obtained; y isjThe jth output value of the hidden layer;
step 4.5, calculating the error signal delta of the output layero：
δo＝[oN(NI-oN)(1-oN)oK(KI-oK)(1-oK)
oV(VI-oV)(1-oV)of(fI-of)(1-of)
ov(vI-ov)(1-ov)]T
Wherein o isNThe output value is the number of the crack strips of the output layer; oKThe output value is the average permeability of the crack strip of the output layer; oVInjecting water waves and volume output values into the output layer; ofInjecting an output value of the water distribution coefficient for the output layer; ovAs output value of water-drive propulsion speed of output layer
Step 4.6, calculating the error signal delta of the hidden layery：
Wherein, deltak oIs the error signal of the output layer; w is aqkThe weight value between the output layer neuron k and the hidden layer neuron q is obtained; w is afwkFor output layer neuron k and hidden layer neuron fwIn betweenA weight value; w is aRCkThe weight value between the output layer neuron k and the hidden layer neuron RC is obtained; w is aRkIs the weight between the output layer neuron k and the hidden layer neuron R; w is aTkThe weight value between the output layer neuron k and the hidden layer neuron T is obtained; w is aNHkThe weight value between the output layer neuron k and the hidden layer neuron NH is obtained; w is aPCkThe weight value between the output layer neuron k and the hidden layer neuron PC is obtained; w is aLkThe weight value between the output layer neuron k and the hidden layer neuron L is obtained; y isqIAn output value for the implicit layer neuron qth sample; y isfwIThe output value for the ith sample of hidden layer neuron fw; y isRCIAn output value for the implicit layer neuron RC sample I; y isRIAn output value for the I < th > sample of hidden layer neuron R; y isTIAn output value for the hidden layer neuron Tth sample; y isPCIs the output value of the hidden layer neuron PC; y isLIs the output value of hidden layer neuron L;
step 4.7, error signal delta of output layer is utilizedoAnd error signal delta of the hidden layeryThe weight is adjusted and corrected by adopting the following formula, namely the weight correction formula of the momentum-adaptive learning rate BP neural network is as follows;
wherein k is 1, 2, …, l; w is ajkThe weight value between the output layer neuron k and the hidden layer neuron j is obtained; j is 1, 2, …, m; v. ofijThe weight value between the input layer neuron i and the hidden layer neuron j is obtained; Δ wjkAnd Δ vijAre respectively the weight value wjkAnd vijChanging the value along the negative gradient direction of the output error E; s is the training times; eta is learning rate, generally 0.01-0.8; mc is a momentum factor, and 0.85-0.95 is taken.
Step 4.8, calculating the error of the BP neural network system:
wherein d isk(I) Outputting the expected output value of the layer neuron k for the I time domain tracing learning sample; ok(I) Outputting the output value of a layer neuron k for the I time domain tracing learning sample;
step 4.9, judging whether the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency;
if the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency, outputting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propulsion speed V of the low-permeability reservoir;
and if the error of the BP neural network system does not meet the condition of the preset precision or exceeds the maximum learning frequency, repeating the step 4.3 to the step 4.9, entering the next round of learning until the error of the BP neural network system meets the condition of the preset precision or exceeds the maximum learning frequency, and finally outputting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the fractured low-permeability oil reservoir.
The time domain tracing neural network test samples and results corresponding to the scheme 1 are shown in table 1:
TABLE 1
Scheme 1 time domain tracing neural network test sample results show that: the absolute error of 4 test samples is 0.269, and the relative error is 4.36%; the maximum absolute error of the number of the crack strips is 0.1, and the relative error is 50 percent; the maximum absolute error of the average permeability of the crack strips is 0.105, and the relative error is 14.1%; the maximum absolute error of injected water wave and volume is 0.121, and the relative error is 108.0%; the maximum absolute error of the distribution coefficient of the injected water is 0.18, and the relative error is 128.6 percent; the maximum absolute error of the water drive propelling speed is 0.23, and the relative error is 35.4%.
When the scheme 2 is taken as an example for solving, the selection of the time domain tracing neural network learning sample and the test sample, the solving method of the model, and the setting of the maximum training times, the minimum training error, the momentum factor and the initial learning rate are all the same as those of the scheme 1. 787 actual training times have an error of 0.67%, and the test results are shown in Table 2:
TABLE 2
Scheme 2 time domain tracing neural network test sample results show that: the absolute error of 4 test samples is 0.120, and the relative error is 1.94%; the maximum absolute error of the number of the crack strips is 0; the maximum absolute error of the average permeability of the crack strips is 0.045, and the relative error is 9.34%; the maximum absolute error of injected water wave and volume is 0.008, and the relative error is 7.08%; the maximum absolute error of the distribution coefficient of the injected water is 0.02, and the relative error is 14.29 percent; the maximum absolute error of the water drive propelling speed is 0.023, and the relative error is 13.37%.
Results of the scheme 1 and the scheme 2 show that the time domain chemical tracing technology and the BP neural network technology can be used for predicting the stratum parameters of the fractured ultra-low permeability reservoir, and the next prediction result can be more accurately judged by considering the previous prediction result when the prediction is performed.
Claims (6)
1. A method for predicting stratum parameters of a fractured ultra-low permeability reservoir is characterized by comprising the following steps:
step 1, establishing a time domain chemical tracing physical model;
step 2, establishing a time domain chemical tracing mathematical model;
step 3, establishing a time domain tracing neural network model;
step 4, solving the time domain chemical tracing mathematical model by using a BP neural network technology according to the time domain chemical tracing physical model, the time domain chemical tracing mathematical model and the time domain tracing neural network model, and predicting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the fractured low-permeability oil reservoir according to the solving result;
the time domain chemical tracing physical model in the step 1 comprises known dynamic history information, known static information, solving variables and a prediction technology, wherein:
known dynamic history information includes: the production time t, the yield, the water content, the well group accumulated injection-production ratio, the well group stage injection-production ratio, the number of the last time of fracture strips, the average permeability of the last time of fracture strips, the last time of water injection swept volume, the last time of water injection distribution coefficient and the last time of water drive propulsion speed;
known static information includes: the static productivity coefficient of the bedrock at the production section, the static energy storage coefficient of the bedrock at the production section and the injection-production well spacing;
solving the variables includes: the number of the crack strips, the average permeability of the crack strips, the swept volume of injected water, the distribution coefficient of the injected water and the water drive propelling speed are controlled;
the prediction technology is a neural network technology;
in the known static information, the static productivity coefficient PC of the bedrock at the production section is equal to the product of the effective permeability k of the bedrock at the production section and the effective thickness h of the bedrock at the production section, and the static energy storage coefficient NH of the bedrock at the production section is equal to the effective thickness h of the bedrock at the production section and the effective porosity of the bedrock at the production section
And oil saturation S of production zone bedrockoThe product of (a);
the single-peak tracer explained physical model is equivalent to an artificial crack channeling model, in the single-peak tracer explained physical model, 1 crack strip is a flow tube bundle consisting of n flow tubes with the length of L and the equivalent diameter of D, tracer flows in the n flow tubes in the crack strip, and the tracer explained model is as follows:
in the formula: c is the output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n is the total number of equivalent flow tubes; vdThe total injected volume of the tracer slug; alpha is hydrodynamic dispersivity; l is the equivalent length of the flow tube; d is the equivalent diameter of the equivalent flow tube; q is the average daily injection;
the multimodal tracer explained physical model is a differential fracture interactive model, a plurality of fracture strips are arranged in the multimodal tracer explained physical model, and the ith fracture strip is distributed from niEach length is LiAn equivalent diameter of DiThe tracer is n in the plurality of fracture stripsiFlow in individual flowtubes, tracer interpretation model is:
in the formula: c is the output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n isiThe number of equivalent flow tubes of the ith crack strip; vdThe total injected volume of the tracer slug; q is the average daily injection; di is the equivalent diameter of any equivalent flow pipe of the ith fracture strip; li is the equivalent flow tube length of the ith crack strip; alpha is alphaiThe hydrodynamic dispersion constant of the tracer in the ith fracture strip equivalent flow tube is obtained;
the physical model of the wide-platform tracer explanation is a relatively uniform crack propulsion type, and is equivalent to a plurality of crack strips, and the ith crack strip is distributed from niEach length is LiAn equivalent diameter of DiThe flow tube bundle formed by the flow tubes, the tracer arrives at the oil well in different flow tubes in sequence, and the tracer interpretation model is as follows:
in the formula: c isThe output concentration of the tracer; c0Is the initial concentration of the tracer; t is time; f. ofjThe distribution coefficient of the injected water to the oil production well j; n isiThe number of equivalent flow tubes of the ith crack strip; vdThe total injected volume of the tracer slug; q is the average daily injection; di is the equivalent diameter of any equivalent flow pipe of the ith fracture strip; li is the equivalent flow tube length of the ith crack strip; alpha is alphaiThe hydrodynamic dispersion constant of the tracer in the ith fracture strip equivalent flow tube is obtained; m is an equivalent flow resistance constant;
when a time domain tracing mathematical model is established, regarding a tracing monitoring process as a time point, and taking historical information monitored by a tracer as a learning sample, wherein the time domain chemical tracing mathematical model is as follows:
in the formula, z is a time domain tracing BP neural network learning sample serial number; n is the number of crack strips; k is the average permeability of the crack band; v is the volume of injected water wave; f is the distribution coefficient of injected water; v is the water drive propulsion speed; q is the yield; f. ofwThe water content is obtained; r is the injection-production ratio of the well group stage; RC is the cumulative injection-production ratio of the well group; w is the weight and the connection coefficient between two adjacent layers.
2. The method for predicting the formation parameters of the fractured ultra-low permeability reservoir of claim 1, wherein in the known dynamic historical information, the number of previous fracture bands, the average permeability of the previous fracture bands, the water swept volume and the volume of the previous injection, the water distribution coefficient of the previous injection and the advancing speed of the previous water drive are calculated by using a fractured ultra-low permeability reservoir tracer classified interpretation model, and the fractured ultra-low permeability reservoir tracer classified interpretation model comprises a unimodal tracer interpreted physical model, a multimodal tracer interpreted physical model or a wide-table tracer interpreted physical model.
3. The method for predicting the formation parameters of a fractured ultra-low permeability reservoir of claim 1, wherein the time-domain tracing neural network model is a three-layer network structure and comprises an input layer, a hidden layer and an output layer group, and the hidden layer is a 1-layer structure; the output layer contains 5 nodes, and 5 nodes are respectively: the number N of the crack strips, the average permeability K of the crack strips, the wave spread volume V of injected water, the distribution coefficient f of the injected water and the water drive propulsion speed V.
4. The method for predicting stratum parameters of a fractured ultra-low permeability reservoir of claim 3, wherein the input layer comprises 8 parameters corresponding to the prediction time, and the 8 parameters are respectively as follows: yield q, water content fwThe method comprises the following steps of (1) well group accumulated injection-production ratio RC, well group stage injection-production ratio R, production time t, injection-production well spacing L, production section bedrock static productivity coefficient PC and production section bedrock static energy storage coefficient NH;
or the input layer contains the following parameters: yield q, water content fwThe method comprises the following steps of well group accumulated injection-production ratio RC, well group stage injection-production ratio R, production time t, injection-production well spacing L, production section bedrock static productivity coefficient PC, production section bedrock static energy storage coefficient NH, last fracture strip number N, last fracture strip average permeability K, last injected water wave volume V, last injected water distribution coefficient f and last water drive propulsion speed V.
5. The method for predicting stratum parameters of fractured ultra-low permeability reservoirs according to claim 4, wherein in the step 4, the process of solving the time domain chemical tracing mathematical model by using the BP neural network technology comprises the following steps:
step 4.1, carrying out normalization processing on the time domain tracing BP neural network learning sample, and setting all weights in the time domain chemical tracing mathematical model as smaller random numbers;
step 4.2, given an input vector X and a desired output vector d: sending the X to an input layer of the time domain tracing neural network model;
4.3, calculating an output vector Y of a hidden layer of the time domain tracing neural network model;
4.4, calculating an output vector O of an output layer of the time domain tracing neural network model;
step 4.5, calculating an error signal delta of an output layer of the time domain tracing neural network modelo；
Step 4.6, calculating an error signal delta of a hidden layer of the time domain tracing neural network modely；
Step 4.7, error signal delta of output layer is utilizedoAnd error signal delta of the hidden layeryAdjusting and correcting the weight;
step 4.8, calculating the error of the BP neural network system;
step 4.9, judging whether the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency;
if the error of the BP neural network system meets the preset precision or exceeds the condition of the maximum learning frequency, outputting the number N of fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propulsion speed V of the low-permeability reservoir;
and if the error of the BP neural network system does not meet the condition of the preset precision or exceeds the maximum learning frequency, repeating the step 4.3 to the step 4.9 until the error of the BP neural network system meets the condition of the preset precision or exceeds the maximum learning frequency, and finally outputting the number N of the fracture strips, the average permeability K of the fracture strips, the injected water wave and volume V, the injected water distribution coefficient f and the water drive propelling speed V of the low-permeability reservoir.
6. The method for predicting stratum parameters of fractured ultra-low permeability reservoir of claim 5, wherein in the step 4.2, the input vector X and the expected output vector d are respectively as follows:
X＝[qI fwI RCI RI TI L PC NH]T
d＝[NI KI VI fI vI]T
wherein q isIYield of learning samples, f, for the ith time domain tracewIWater content of learning sample for I time domain traceRate, RCICumulative injection-production ratio, R, for the well group of the I time domain tracer learning samplesIWell group stage injection-production ratio, T, for the I time domain tracer learning sampleITracing the production time of the learning sample for the I time domain, wherein L is the injection-production well spacing, PC is the static productivity coefficient of the bedrock of the production section, NH is the static energy storage coefficient of the bedrock of the production section, and N isINumber of crack bands, K, for the I time domain trace learning sampleIAverage permeability of crack bands, V, for the I time domain trace learning sampleIVolume of injected water wave, f, for the I time domain tracer learning sampleIDistribution coefficient of injected water, v, for the I time domain tracer learning sampleITracing the water drive propulsion speed of the learning sample for the I time domain;
in step 4.3, the output vector Y is:
wherein A is a threshold value of each neuron of the hidden layer;
in step 4.4, the output vector O is:
b is a threshold value of each neuron of an output layer, and B is a threshold value of each neuron of the output layer; w is ajNThe weight value between the output layer neuron N and the hidden layer neuron j is obtained; w is ajKThe weight value between the output layer neuron K and the hidden layer neuron j is obtained; w is ajVThe weight value between the output layer neuron V and the hidden layer neuron j is obtained; w is ajfThe weight value between the output layer neuron f and the hidden layer neuron j is obtained; w is ajvThe weight value between the output layer neuron v and the hidden layer neuron j is obtained; y isjThe jth output value of the hidden layer;
in step 4.5, the error signal delta of the layer is outputoComprises the following steps:
δo＝[oN(NI-oN)(1-oN) oK(KI-oK)(1-oK) oV(VI-oV)(1-oV) of(fI-of)(1-of) ov(vI-ov)(1-ov)]T
wherein o isNThe output value is the number of the crack strips of the output layer; oKThe output value is the average permeability of the crack strip of the output layer; oVInjecting water waves and volume output values into the output layer; ofInjecting an output value of the water distribution coefficient for the output layer; ovThe output value is the water drive propulsion speed of the output layer;
in step 4.6, the error signal δ of the hidden layeryComprises the following steps:
wherein, deltak oIs the error signal of the output layer; w is aqkThe weight value between the output layer neuron k and the hidden layer neuron q is obtained; w is afwkFor output layer neuron k and hidden layer neuron fwThe weight value between; w is aRCkThe weight value between the output layer neuron k and the hidden layer neuron RC is obtained; w is aRkIs the weight between the output layer neuron k and the hidden layer neuron R; w is aTkThe weight value between the output layer neuron k and the hidden layer neuron T is obtained; w is aNHkThe weight value between the output layer neuron k and the hidden layer neuron NH is obtained; w is aPCkThe weight value between the output layer neuron k and the hidden layer neuron PC is obtained; w is aLkThe weight value between the output layer neuron k and the hidden layer neuron L is obtained; y isqIAn output value for the implicit layer neuron qth sample; y isfwIThe output value for the ith sample of hidden layer neuron fw; y isRCIAn output value for the implicit layer neuron RC sample I; y isRIAn output value for the I < th > sample of hidden layer neuron R; y isTIAn output value for the hidden layer neuron Tth sample; y isPCIs the output value of the hidden layer neuron PC; y isLTo hide the spirit of the layerAn output value via element L;
in step 4.7, the error signal delta of the output layer is usedoAnd error signal delta of the hidden layeryAnd (3) adjusting and correcting the weight by adopting the following formula:
wherein, output layer neuron k is 1, 2, …, l; w is ajkThe weight value between the output layer neuron k and the hidden layer neuron j is obtained; output layer neuron j ═ 1, 2, …, m; v. ofijThe weight value between the input layer neuron i and the hidden layer neuron j is obtained; Δ wjkAnd Δ vijAre respectively the weight value wjkAnd vijChanging the value along the negative gradient direction of the output error E; s is the training times; eta is the learning rate, and the value range is 0.01-0.8; mc is a momentum factor, and the value range is 0.85-0.95;
in step 4.8, the error of the BP neural network system is:
wherein d isk(I) Outputting the expected output value of the layer neuron k for the I time domain tracing learning sample; ok(I) And outputting the output value of the layer neuron k for the I time domain tracing learning sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810996052.7A CN109209361B (en) | 2018-08-29 | 2018-08-29 | Method for predicting stratum parameters of fractured ultra-low permeability reservoir |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810996052.7A CN109209361B (en) | 2018-08-29 | 2018-08-29 | Method for predicting stratum parameters of fractured ultra-low permeability reservoir |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109209361A CN109209361A (en) | 2019-01-15 |
| CN109209361B true CN109209361B (en) | 2022-02-18 |
Family
ID=64985286
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810996052.7A Active CN109209361B (en) | 2018-08-29 | 2018-08-29 | Method for predicting stratum parameters of fractured ultra-low permeability reservoir |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109209361B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800521A (en) * | 2019-01-28 | 2019-05-24 | 中国石油大学(华东) | A kind of oil-water relative permeability curve calculation method based on machine learning |
| CN113323618A (en) * | 2021-06-02 | 2021-08-31 | 德仕能源科技集团股份有限公司 | Ultra-low permeability reservoir artificial fracture water channeling treatment method and device |
| CN115961923B (en) * | 2021-10-12 | 2024-10-18 | 中国石油天然气股份有限公司 | Deep profile control and flooding method for fractured reservoir |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106526693A (en) * | 2016-09-30 | 2017-03-22 | 中国石油天然气股份有限公司 | crack identification method and device |
| CN107989597A (en) * | 2017-11-15 | 2018-05-04 | 中国石油大学(北京) | A kind of crack data screening technique, device and storage medium |
| CN108301823A (en) * | 2018-01-19 | 2018-07-20 | 北京捷贝通石油技术股份有限公司 | A method of identification reservoir hydrocarbons dessert |
| CN108319756A (en) * | 2017-12-29 | 2018-07-24 | 西安石油大学 | A kind of compact reservoir volume fracturing seam net extended simulation and characterizing method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140025301A1 (en) * | 2012-07-20 | 2014-01-23 | Bruce H. Storm, Jr. | Determination of subsurface properties of a well |
| US10599987B2 (en) * | 2016-07-14 | 2020-03-24 | King Fahd University Of Petroleum And Minerals | Apparatuses, systems, and methodologies for permeability prediction |
-
2018
- 2018-08-29 CN CN201810996052.7A patent/CN109209361B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106526693A (en) * | 2016-09-30 | 2017-03-22 | 中国石油天然气股份有限公司 | crack identification method and device |
| CN107989597A (en) * | 2017-11-15 | 2018-05-04 | 中国石油大学(北京) | A kind of crack data screening technique, device and storage medium |
| CN108319756A (en) * | 2017-12-29 | 2018-07-24 | 西安石油大学 | A kind of compact reservoir volume fracturing seam net extended simulation and characterizing method |
| CN108301823A (en) * | 2018-01-19 | 2018-07-20 | 北京捷贝通石油技术股份有限公司 | A method of identification reservoir hydrocarbons dessert |
Non-Patent Citations (2)
| Title |
|---|
| 四维试井理论研究;吴洪彪 等;《石油学报》;20030930;第24卷(第5期);57-62页 * |
| 裂缝性特低渗透油藏水窜水淹逐级调控多级井间化学示踪技术;蒲春生 等;《石油勘探与开发》;20160823;第43卷(第4期);621-629页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109209361A (en) | 2019-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109209361B (en) | Method for predicting stratum parameters of fractured ultra-low permeability reservoir | |
| Panda et al. | An integrated approach to estimate well interactions | |
| CN105117511B (en) | A kind of characterizing method of fracture hole oil reservoir interwell communication passage and flow parameter | |
| CN108984886B (en) | Method for inverting dynamic connectivity among wells of multi-layer oil reservoir | |
| Huang et al. | Comparison of different machine learning algorithms for predicting the SAGD production performance | |
| CN101725346A (en) | Oil deposit inter-well dynamic connectivity inverting method | |
| Ramgulam et al. | Utilization of artificial neural networks in the optimization of history matching | |
| Du et al. | The connectivity evaluation among wells in reservoir utilizing machine learning methods | |
| CN109800521A (en) | A kind of oil-water relative permeability curve calculation method based on machine learning | |
| CN105386751A (en) | Well logging and productivity prediction method of horizontal well based on oil reservoir seepage flow model | |
| CN114781951B (en) | Shale oil reservoir carbon dioxide huff-puff development well selection decision method and system | |
| CN109063403B (en) | Optimal design method for slickwater fracturing | |
| CN112539054B (en) | Production optimization method for complex system of ground pipe network and underground oil reservoir | |
| CN107895092A (en) | A kind of interwell communication quantitative evaluation method that modeling is adopted based on complex nonlinear note | |
| CN113762559A (en) | Intelligent yield splitting method and system for multilayer commingled production gas well | |
| CN106481315A (en) | Land sandstone oil reservoir individual well recoverable reserves quickly determines model and method for building up | |
| CN110486008A (en) | A kind of parameter interpretation method and system of Radial Compound Reservoir | |
| CN110991084A (en) | A Reservoir Permeability Calculation Method Based on Streamline Numerical Well Testing | |
| CN112464136B (en) | Method for predicting directional well production energy and development effect of offshore thin interbed sandstone oil field | |
| CN118246579A (en) | Mixed rock multi-mineral volume content prediction method and device based on deep learning | |
| Temizel et al. | Production Forecasting in Conventional Oil Reservoirs Using Deep Learning | |
| CN113837482A (en) | Dynamic prediction method for solution-fractured reservoir | |
| Zhu et al. | Calculation method of production pressure drop based on BP neural network velocity pipe string production in CBM wells | |
| Liu et al. | A robust method to predict fluid properties based on big data and machine learning algorithms | |
| Isaiah et al. | Performing reservoir simulation with Neural Network enhanced data |
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 |