CN103235879A - Bi-phase medium parametric inversion method based on niche master-slave parallel genetic algorithm - Google Patents

Abstract

The invention relates to a bi-phase anisotropic medium reservoir parametric inversion method based on a niche master-slave parallel genetic algorithm. According to the method, the niche master-slave parallel genetic algorithm is used for solving bi-phase anisotropic medium reservoir parameters, the core ideology includes that a system comprises a master processor and a plurality of slave processors, the master processor monitors a whole population, at a fitness calculation stage, the master processor distributes calculation of the fitness to all slave processors, collects results after calculation and then performs operations such as niche elimination, selection, cross and variation to generate a new generation of population so as to finish one circulation, and the calculation efficiency of reservoir parametric inversion is improved greatly. According to the method, a concept of sharing degree is introduced in the reservoir parameter evolution solving process, substantial growth of some individuals are limited through adjustment of the fitness of each individual, niche evolution environments are created, and the capacity for solving multiple-peak reservoir parametric inversion optimization problems and the solving quality through the genetic algorithm are improved. The bi-phase medium parametric inversion method is widely applied to parametric inversion processes of oil and gas reservoirs.

Description

Two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm

Technical field

The present invention relates to the inversion method of reservoir parameter in a kind of petroleum exploration field, particularly about a kind of two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm.

Background technology

The inverting of reservoir parameter (as factor of porosity, stream phase density and solid phase density) is significant for the dynamic monitoring in the oil and gas development, optimum management.Along with deepening continuously of exploration of oil and gas field and exploitation degree, the problem that we face also becomes increasingly complex, particularly the reservoir problem.Generally speaking, the occurrence form of hydrocarbon-bearing pool is crack, crack or pore type, so hydrocarbon-bearing pool often shows as two-phase media, and traditional pure solid elastic dielectrics theory is incomplete, and description rock that can not be detailed is to the influence of ripple communication process.Because the two-phase media theory can be described actual formation structure and character more accurately, so also just more can adapt to the actual needs of the reservoir exploration that becomes increasingly complex.

In the research process of two-phase media ^ THE THEORY OF ELASTIC WAVE, the Biot(name) set up the communication theory of ripple in solid-stream coupling two-phase pore media, established the basis of two-phase media elastic wave theory, but the Biot theory can not be explained overdamp and the high dispersion phenomenon of ripple well, Dvorkin and Nur combined the Biot flow mechanism by introducing feature injection stream length with injection (Squirt) flow mechanism afterwards, realized the combination of macro-scale and micro-scale, proposed to contain Biot-Squirt(BISQ unified in the saturated with fluid pore media) model.Because Biot flows and jet flow is to contain the mobile topmost two kinds of liquid forms of fluid in the fluid reservoir medium, the BISQ model that takes into full account this two kinds of hydromechanicses mechanism and set up can be truer, the propagation law of describing reservoir medium medium wave exactly, so realize having important practical significance for the practical problems that solves oil-gas exploration and exploitation based on the double phase anisotropic medium reservoir parameter inverting of BISQ model.At present, in general two-phase media reservoir parameter inversion method mainly is divided into two classes: the first kind is based on isotropy Biot model, and its inverting is according to being dielectric surface displacement response; Second class is based on isotropy or anisotropy BISQ model, and its inverting foundation is phase velocity.Working medium surface displacement response of the present invention has realized the reservoir parameter inverting based on BISQ model double phase anisotropic medium as the inverting foundation for this reason.Should the principle consistent with actual measurement data according to the theoretical synthetic displacement response of dielectric surface, two-phase media parametric inversion problem finally can be summed up as the minimum problem of non-linear functional multimodal function.Yet the optimization of non-linear functional multimodal function is a very difficult problem.This is because if adopt traditional optimization method, as gradient method, Newton method etc., all need to find the solution Jacobi matrix or extra large gloomy matrix etc., for the two-phase media model, this solution procedure is very complicated, and if will be applied in the practical problems also very difficult; Simultaneously, adopt the inversion result of traditional optimization obviously to depend on choosing of initial point, be difficult to obtain globally optimal solution.Genetic algorithm is by the process of simulation biological heredity and evolution, utilizes the transition probability rule to help guidance search, and Search Results does not rely on choosing of initial point, has very strong robustness for finding the solution globally optimal solution; Simultaneously, because the control information of only utilizing objective function to bring makes to calculate to become easy relatively.Niche genetic algorithm is as a kind of improvement algorithm of conventional genetic algorithm, can keep the diversity of separating well, have characteristics such as preventing " precocity ", minimizing disturbance, this is the relative advantage place of other algorithms in the research of the non-linear multiparameter inverting of geophysical survey optimization problem of this algorithm just.Yet at two-phase media reservoir parameter inverse problem, the fitness calculated amount of microhabitat serial genetic algorithm is very big, so its optimization efficient is very low, and inverting is very consuming time.

Summary of the invention

At the problems referred to above, the purpose of this invention is to provide and a kind ofly can effectively improve reservoir parameter inverting solving precision, accelerate counting yield significantly, based on the two-phase media parameter inversion method of microhabitat master-slave mode paralleling genetic algorithm.

For achieving the above object, the present invention takes following technical scheme: a kind of two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm may further comprise the steps: 1) according to calculation requirement, carry out building of MPI parallel computing platform; According to calculation requirement, MPICH is installed under the Windows/Linux system and disposes, carry out building of MPI parallel computing platform; 2) according to actual geologic information, make two-phase media BISQ model, as the implementation model of microhabitat master-slave mode paralleling genetic algorithm reservoir parameter inverting; 3) according to principle of least square structure objective function:

$E\left(p\right)=\frac{1}{2}{\left|\right|A\left(p\right)-G^{*}\left|\right|}^2,$

Wherein || || be the L-2 norm, p=[ρ _f, ρ _s, φ] ^-TWait to ask parameter stream phase density ρ for discretize _f, the solid phase density p _sWith the one-dimensional vector that factor of porosity φ forms in a certain order, A is a vector valued function A:p → G, and G is the synthetic x component of theory of face of land medium solid phase displacement response and the one-dimensional vector that the z component forms according to a definite sequence, G ^*Be the actual measurement data x component and the one-dimensional vector of z component according to the order formation identical with G of face of land medium solid phase displacement response; 4) the main operational factor of setting genetic algorithm comprises population scale N, crossover probability P _c, the variation probability P _m, maximum genetic evolution algebraically GEN _Max, the minimum iteration error error during convergence and each inverted parameters span; 5) open MPI parallel computing platform MPICH2, the algorithm operational factor in the step 4) is loaded into, Num process opened by system, comprise a host process and Num-1 from process, initialization of population utilizes the floating-point encoding method to produce initial population by primary processor, and population scale is N; 6) by primary processor according to centralized dispatching dynamic task allocation strategy divide mating group to each from processor, carry out fitness according to the objective function of described step 3) structure and calculate, treat all from the processor fitness calculate finish after again by primary processor collection result; 7) use Sharing Function to adjust each individual fitness in the population, and in the population evolutionary process after this, the new fitness after algorithm is adjusted according to this is selected to eliminate computing with microhabitat; 8) fitness that at first step 7) calculated of primary processor carries out the conversion of sigma cut-off scales and obtains new fitness, according to this fitness population is not had the playback remainder then and selects computing at random, and adopt optimum conversation strategy, obtains body set one by one; 9) primary processor individual collections that step 8) is selected is done simulation scale-of-two crossing operation; 10) primary processor carries out the non-uniform mutation computing to the population that the step 9) crossing operation obtains, and generates progeny population; 11) step 10) is made a variation result that computing obtains carries out the reservoir parameter Inversion Calculation according to the objective function of step 3) structure, if meet end condition, then enter next step, continue to calculate as initial population otherwise the result that the computing that will make a variation obtains turns back to step 6); 12) will satisfy the result of calculation of end condition as the net result output of BISQ model double phase anisotropic medium reservoir parameter inverting, EOP (end of program).

Centralized dispatching dynamic task allocation strategy in the described step 6) refers to that system is divided into a primary processor and some from processor, the whole population of main processor monitors, calculation stages at fitness, by primary processor the dispensed of fitness is carried out from processor to each, when one when idle from processor, tell primary processor by message, primary processor will take out one by one body and distribute to this and proceed fitness from processor and calculate from the remaining individuality of population, fitness calculating up to whole population finishes, and collects the result by primary processor again.

In the described step 6), carry out fitness when calculating according to the objective function of described step 3) structure, at first will obtain the face of land medium solid phase displacement response of actual measurement as the inverting foundation, then at the parameter solid phase density p of wanting inverting _s, stream phase density ρ _fWith factor of porosity φ, two-phase media reservoir parameter inverse problem is considered as p (ρ in three dimensions _f, ρ _s, φ) seek 1 p ^*(ρ _f ^*, ρ _s ^*, φ ^*), make the theory of its corresponding face of land medium solid phase displacement response synthesize the face of land medium solid phase displacement response that under the least square meaning, is fit to actual measurement best, at last according to principle of least square structure objective function.

In the described step 7), Sharing Function is a function of substantial connection degree between two individualities in the expression colony, with sh (d _Ij) expression, the fitness of the individuality after adjusting is that its former fitness value is divided by sharing degree m _i, namely

F'(X _i)＝F(X _i)/m _i

Wherein, F (X _i) be individual X _iFormer fitness value; m _iBe the Sharing Function sh (d between other individualities in individual i and the colony _Ij) sum, represent a kind of tolerance of individual i degree of share in colony; m _iExpression formula as follows:

$m_i=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{sh}\left(d_{\mathrm{ij}}\right)$ (i＝1,2,…,N)。

In the described step 8), the conversion of described Sigma cut-off scales be with original fitness function F hint obliquely at for , wherein c is a very little integer, and general value is c ∈ [1,5], and the value of c makes all fitness all more than or equal to 0,

With σ be respectively mean value and the variance of population fitness; The specific operation process that described no playback remainder is selected at random is: at first calculate the existence desired number M of each individuality in colony of future generation in the colony _i,

(i=1,2 ..., N).Get M then _iIntegral part [M _i] be corresponding individual existence number in colony of future generation, can determine so altogether in the N of future generation colony

Individuality.At last with

Be the new fitness of each individuality, use and determine in the colony of future generation also undetermined at random than case selection method

Individuality, described optimum conversation strategy is that the highest fitness value of each colony of new generation that produces after for genetic manipulation and the highest fitness value of previous generation colony are made comparisons, if the highest adaptive value less than previous generation, just eliminate the body one by one in a new generation at random, the individuality of high adaptive value joins in a new generation having among the previous generation.

In the described step 9), the method for operating when doing simulation scale-of-two crossing operation is to the individual x of two fathers ₁And x ₂, generate two individual c of son in such a way ₁And c ₂:

$\left\{\begin{array}{c}{c}_{1,i}=[(1+\mathrm{\β})x_{1,i}+(1-\mathrm{\β})x_{2,i}]/2\\ c_{2,i}=[(1-\mathrm{\β})x_{1,i}+(1+\mathrm{\β})x_{2,i}]/2\end{array}\right.$ ，1≤i≤n

Wherein β is stochastic variable, all needs to regenerate on each dimension, and mode is as follows:

$\mathrm{\β}=\left\{\begin{array}{cc}{\left(2\mathrm{\μ}\right)}^{\frac{1}{\mathrm{\η}+1}}& u\≤0.5\\ {\left(2(1-\mathrm{\μ})\right)}^{-\frac{1}{\mathrm{\η}+1}}& u>0.5\end{array}\right.$

In the formula, μ is the random number that is uniformly distributed on the interval (0,1), and η is cross parameter, is an integer.

In the described step 10), generate in the process of progeny population, if carrying out by x=x ₁x ₂X _kX _nTo x'=x ₁x ₂X _k' ... x _nNon-uniform mutation when operation, if change point x _kThe genic value span at place is

New genic value x then _k' determined by following formula:

Wherein Δ (t, y) (y representative

With

) expression meets a random number of non-uniform Distribution in [0, y] scope, requires the increase along with evolutionary generation t, (t y) also increases close to 0 probability Δ gradually.

The end condition of described step 11) refers to that objective function satisfies trueness error error requirement or genetic evolution algebraically reaches maximal value GEN _Max

The present invention is owing to take above technical scheme, it has the following advantages: 1, two-phase media BISQ model theory used in the present invention can be described the fluid properties on actual formation architectural characteristic and stratum exactly, is more suitable in the actual needs of disguised reservoir exploration and oil-field development than single-phase medium theory of elasticity.2, the present invention according to the theory of dielectric surface displacement response synthetic should with the principle of actual measurement data phase match, introduce niche genetic algorithm and master-slave mode parallel calculating method, set up the microhabitat master-slave mode parallel computation multiparameter joint inversion method based on BISQ model two-phase media.With respect to the conventional genetic algorithm of prior art, the inventive method not only can effectively improve inversion accuracy, but also can accelerate counting yield significantly.3, the present invention eliminates the concept of introducing sharing degree in the computing in microhabitat, by adjusting the dramatic growth that each individual fitness limits individual one, bring up the evolution environment of microhabitat, improve genetic algorithm and handled the ability of multi-peak reservoir parameter inverting optimization problem, thereby greatly promoted the quality of finding the solution of genetic algorithm for solving two-phase media reservoir parameter inverse problem.4, the present invention combines high-speed parallel and the intrinsic concurrency of niche genetic algorithm of parallel computer, form microhabitat master-slave mode paralleling genetic algorithm, the core concept of this algorithm, system is divided into a primary processor and some from processor, the whole population of main processor monitors, calculation stages at fitness, by primary processor the dispensed of fitness is carried out from processor to each, collect the result by primary processor again after calculating is finished and be used for subsequent calculations, can accelerate counting yield significantly like this.The present invention has certain actual application value and good prospects for application for the dynamic monitoring in the oil and gas development and optimum management.

Description of drawings

Fig. 1 is microhabitat master-slave mode paralleling genetic algorithm FB(flow block) of the present invention

Fig. 2 is that inverted parameters of the present invention and objective function are with the variation tendency of genetic algebra

Fig. 3 is that the conventional genetic inverse parameter of prior art and objective function are with the variation tendency of genetic algebra

Fig. 4 is that prior art microhabitat serial genetic inverse parameter and objective function are with the variation tendency of genetic algebra

Embodiment

Below in conjunction with drawings and Examples the present invention is described in detail.

The present invention is based on the double phase anisotropic medium reservoir parameter inversion method of microhabitat master-slave mode paralleling genetic algorithm, may further comprise the steps:

1) according to calculation requirement, carries out building of MPI parallel computing platform.MPICH2 is the MPI(message passing interface) most important a kind of specific implementation, according to calculation requirement, MPICH2 is installed and configuration under the Windows7 system and among the Visual Studio2008, carry out building of MPI parallel computing platform;

2) with reference to actual geologic information, make and survey district's target reservoir geologic model.According to well logging and rock physics data, make two-phase media BISQ model, as the implementation model of microhabitat master-slave mode paralleling genetic algorithm reservoir parameter inverting;

3) determining stream phase density ρ _f, the solid phase density p _sUnder factor of porosity φ condition, find the solution the BISQ model face of land solid phase displacement response x component u that has set up by the staggered-mesh method of finite difference _x(x, 0, t) with z component u _z(x, 0, t) simulate the face of land medium solid phase displacement response of actual measurement, suppose that then solid phase density, stream phase density and factor of porosity are parameter to be asked, and can be considered as p (ρ in three dimensions to two-phase media reservoir parameter inverse problem like this _f, ρ _s, φ) seek 1 p ^*(ρ _f ^*, ρ _s ^*, φ ^*), make the theory of its corresponding face of land medium solid phase displacement response synthesize the face of land medium solid phase displacement response that under the least square meaning, is fit to actual measurement best.

Construct objective function according to the principle of least square:

$E\left(p\right)=\frac{1}{2}{\left|\right|A\left(p\right)-G^{*}\left|\right|}^2,$

Wherein || || be the L-2 norm, p=[ρ _f, ρ _s, φ] ^TWait to ask parameter stream phase density ρ for discretize _f, the solid phase density p _sWith the one-dimensional vector that factor of porosity φ forms in a certain order, A is a vector valued function A:p → G, and G is the synthetic x component of theory of face of land medium solid phase displacement response and the one-dimensional vector that the z component forms according to a definite sequence, G ^*Be the actual measurement data x component and the one-dimensional vector of z component according to the order formation identical with G of face of land medium solid phase displacement response; Make objective function E (p) obtain minimizing vectorial p ^*Be inverse problem and wait to ask the solution of parameter;

4) the main operational factor of setting genetic algorithm comprises population scale N, crossover probability P _c, the variation probability P _m, maximum genetic evolution algebraically GEN _Max, the minimum iteration error error during convergence and each inverted parameters span; This process need is selected suitable genetic algorithm operational factor through experiment test, just can obtain inversion result preferably; Based on above parameter, will carry out the parametric inversion of BISQ model two-phase media according to microhabitat master-slave mode paralleling genetic algorithm process flow diagram shown in Figure 1 below;

5) open MPI parallel computing platform MPICH2, algorithm operational factor in the step 4) is loaded into, Num process opened by system, individual from process comprising a host process and Num-1, initialization population then, utilize the floating-point encoding method to produce initial population (set of initial solution) by primary processor, population scale is N;

6) by primary processor according to centralized dispatching dynamic task allocation strategy divide mating group to each from processor, each objective function of constructing according to step 3) from processor carries out fitness and calculates, when one when idle from processor, tell primary processor by message, primary processor will take out one by one body and distribute to this and proceed fitness from processor and calculate from the remaining individuality of population, calculate up to the fitness of whole population and to finish, treat all from the processor fitness calculate finish after again by primary processor collection result;

7) use Sharing Function to adjust each individual fitness in the population, the dramatic growth of restriction individual one, in the population evolutionary process after this, new fitness after algorithm is adjusted according to this is selected to eliminate computing with microhabitat, to safeguard the diversity of population, create the evolution environment of microhabitat.

Sharing Function is a function of substantial connection degree between two individualities in the expression colony, with sh (d _Ij) expression, the fitness of the individuality after adjusting is that its former fitness value is divided by sharing degree m _i, namely

F'(X _i)＝F(X _i)/m _i

Wherein, F (X _i) be individual X _iFormer fitness value; m _iBe the Sharing Function sh (d between other individualities in individual i and the colony _Ij) sum, represent a kind of tolerance of individual i degree of share in colony; m _iExpression formula as follows:

$m_i=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{sh}\left(d_{\mathrm{ij}}\right)$ (i＝1,2,…,N)

8) primary processor at first carries out sigma(σ to the fitness that step 7) is calculated) the cut-off scales conversion obtains new fitness, according to this fitness population is not had the playback remainder then and select computing at random, and adopt optimum conversation strategy, obtain body set one by one; The conversion of Sigma cut-off scales be with original fitness function F hint obliquely at for

Wherein c is a very little integer, and general value is c ∈ [1,5], and the value of c makes all fitness all more than or equal to 0,

With σ be respectively mean value and the variance of population fitness.

The specific operation process that no playback remainder is selected at random is: at first calculate the existence desired number M of each individuality in colony of future generation in the colony _i,

Get M then _iIntegral part [M _i] be corresponding individual existence number in colony of future generation, can determine so altogether in the N of future generation colony

Individuality.At last with

Be the new fitness of each individuality, use and determine in the colony of future generation also undetermined at random than case selection method

Individuality.This selection method of operating can guarantee that fitness some individualities one bigger than average fitness are genetic in the colony of future generation surely, and Select Error is smaller.

Optimum conversation strategy is that the highest fitness value of each colony of new generation that produces after for genetic manipulation and the highest fitness value of previous generation colony are made comparisons, if the highest adaptive value less than previous generation, just eliminate the body one by one in a new generation at random, the individuality of high adaptive value joins in a new generation having among the previous generation.Can guarantee that so current optimum individual can not destroyed by hereditary computings such as intersection, variations, be an important guaranteed conditions of genetic algorithm converges;

9) primary processor individual collections that step 8) is selected is done simulation scale-of-two crossing operation; Its method of operating is to the individual x of two fathers ₁And x ₂, generate two individual c of son in such a way ₁And c ₂:

$\left\{\begin{array}{c}{c}_{1,i}=[(1+\mathrm{\β})x_{1,i}+(1-\mathrm{\β})x_{2,i}]/2\\ c_{2,i}=[(1-\mathrm{\β})x_{1,i}+(1+\mathrm{\β})x_{2,i}]/2\end{array}\right.$ ，1≤i≤n

Wherein β is stochastic variable, all needs to regenerate on each dimension, and mode is as follows:

$\mathrm{\β}=\left\{\begin{array}{cc}{\left(2\mathrm{\μ}\right)}^{\frac{1}{\mathrm{\η}+1}}& u\≤0.5\\ {\left(2(1-\mathrm{\μ})\right)}^{-\frac{1}{\mathrm{\η}+1}}& u>0.5\end{array}\right.$

In the formula, μ is the random number that is uniformly distributed on the interval (0,1), and η is cross parameter, is an integer.

10) primary processor carries out the non-uniform mutation computing to the population that the step 9) crossing operation obtains, and generates progeny population; If carrying out by x=x ₁x ₂X _kX _nTo x'=x ₁x ₂X _k' ... x _nNon-uniform mutation when operation, if change point x _kThe genic value span at place is

, new genic value x then _k' determined by following formula:

Wherein Δ (t, y) (y representative With ) expression meets a random number of non-uniform Distribution in [0, y] scope, requires the increase along with evolutionary generation t, (t y) also increases close to 0 probability Δ gradually.

11) step 10) is made a variation result that computing obtains carries out the reservoir parameter Inversion Calculation according to the objective function of step 3) structure, and (objective function satisfies that trueness error error requires or genetic evolution algebraically reaches maximal value GEN if meet end condition _Max), then stop calculating and the output inversion result, otherwise turning back to step 6), the result that the computing that will make a variation obtains continues to calculate till satisfying end condition as initial population.

12) result of calculation that step 11) is exported is as the net result of BISQ model double phase anisotropic medium reservoir parameter inverting, and EOP (end of program) is calculated.

For the effect of above-mentioned embodiment better is described, provide an instantiation below:

According to calculation requirement, MPICH2 is installed and configuration under the Windows7 system and among the Visual Studio2008, carry out building of MPI parallel computing platform.

According to well logging and rock physics data, make two-dimentional semispace two-phase fissuted medium BISQ model, as the implementation model (as shown in table 1) of microhabitat master-slave mode paralleling genetic algorithm reservoir parameter inverting.

Table 1: model parameter

The place of measuring seismologic record is chosen at table 1 model surface, i.e. the z=0 place.Obtain this two-phase media model face of land solid phase displacement response u by the staggered-mesh method of finite difference _x=(x, 0, t) and u _z=(x, 0, t) as the subsidiary condition of parametric inversion, namely Shi Ji seismic measurement data is supposed the parameter phi of wanting inverting, ρ again _sAnd ρ _fBe unknown, the structure objective function adopts microhabitat master-slave mode paralleling genetic algorithm to carry out numerical inversion at last.

The main operational factor of setting genetic algorithm is: population population size N=20; Crossover probability P _c=0.9; The variation probability P _m=0.15; Maximum genetic evolution algebraically GEN _Max=100; Minimum iteration error error=10 during convergence ^-40Each inverted parameters span is respectively: φ (0.1-0.5), ρ _s(2000-3000kg/m ³) and ρ _f(500-1500kg/m ³).

Based on above parameter, under following test environment (as shown in table 2), use three processors (actual numerical value is more than processor number used herein when calculating, and has only conveniently used less processor in order to say something herein) to carry out the BISQ model two-phase fissuted medium reservoir parameter inverting of microhabitat master-slave mode paralleling genetic algorithm far away according to microhabitat master-slave mode paralleling genetic algorithm process flow diagram.The refutation process analysis as shown in Figure 2, after inverting finishes, inversion result and the error of each reservoir parameter following (as shown in table 3):

Table 2: performance detection environment

Table 3: microhabitat master-slave mode paralleling genetic algorithm reservoir parameter inversion result

Inverted parameters	Factor of porosity φ	The solid phase density p s(kg/m 3)	Stream phase density ρ f(kg/m 3)
				Theoretical value	0.25	2600	1000
Inverting value	0.2502	2591.9	981·8
				Relative error (%)	0.08	0.3115	1.82

As shown in Figure 2, along with the iteration of genetic algorithm, the parameter phi of microhabitat master-slave mode paralleling genetic algorithm inverting, ρ _sAnd ρ _fCan converge to their optimum solution (theoretical value shown in the solid line) separately rapidly, the objective function speed of convergence is also very fast.Associative list 3 can draw, and the inversion result of three reservoir parameters of microhabitat master-slave mode paralleling genetic algorithm and model true value are coincide finely, and inversion accuracy is higher, and its maximum error is 1.82%, and least error only is 0.08%.Consuming timely be about 56.37 hours for 100 generations of this model microenvironment master-slave mode paralleling genetic algorithm inverting.

Based on identical operational factor and the test environment of above-mentioned microhabitat master-slave mode paralleling genetic algorithm, use a processor to carry out the BISQ model two-phase fissuted medium reservoir parameter inverting of the conventional genetic algorithm of prior art, the refutation process analysis as shown in Figure 3, inversion result and the error of each reservoir parameter following (as shown in table 4):

Table 4: conventional genetic algorithm reservoir parameter inversion result

Inverted parameters	Factor of porosity φ	The solid phase density p s(kg/m 3)	Stream phase density ρ f(kg/m 3)
				Theoretical value	0.25	2600	1000
Inverting value	0.2512	2526.2	836.0
				Relative error (%)	0.48	2.8385	16.4

As shown in Figure 3, as can be seen from the figure, along with the iteration of genetic algorithm, the parameter phi of the conventional genetic inverse of prior art, ρ _sAnd ρ _fDo not converge to their optimum solutions separately, and the inversion result disturbance is violent in the genetic evolution solution procedure.Associative list 4 can draw, and the conventional genetic inverse ratio of precision of prior art is relatively poor, and error ratio is bigger.And consuming timely being about 110.36 hours for 100 generations of the conventional genetic inverse of same model prior art, approximately is microhabitat master-slave mode paralleling genetic algorithm inverting of the present invention twice consuming time.

Based on identical operational factor and the test environment of above-mentioned microhabitat master-slave mode paralleling genetic algorithm, use a processor to carry out the BISQ model two-phase fissuted medium reservoir parameter inverting of prior art microhabitat serial genetic algorithm, the refutation process analysis as shown in Figure 4, inversion result and the error of each reservoir parameter following (as shown in table 5):

Table 5: microhabitat serial genetic algorithm reservoir parameter inversion result

Inverted parameters	Factor of porosity φ	The solid phase density p s(kg/m 3)	Stream phase density ρ f(kg/m 3)
				Theoretical value	0.25	2600	1000
Inverting value	0.2502	2591.9	981.8
				Relative error (%)	0.08	0.3115	1.82

As shown in Figure 4, as can be seen from the figure, along with the iteration of genetic algorithm, the parameter phi of microhabitat serial genetic inverse, ρ _sAnd ρ _fCan converge to their optimum solution separately rapidly, the objective function speed of convergence is also very fast, and can reduce disturbing phenomenon effectively.Associative list 5 can draw, the conventional genetic algorithm of relative prior art, the inversion result of three reservoir parameters of prior art microhabitat serial genetic algorithm and model true value are coincide better, and inversion accuracy is higher, its maximum error is 1.82%, and least error only is 0.08%.Consuming timely be about 110.47 hours for 100 generations of same model prior art microhabitat serial genetic inverse, as can be seen consuming time from the reservoir parameter inverting, with respect to the conventional genetic algorithm of prior art, prior art microhabitat serial genetic algorithm has just obtained better inversion result in the time of only need increasing the computing machine of minute quantity.

From Fig. 2 and Fig. 4 and table 3 and table 5 as can be seen, the result that microhabitat master-slave mode paralleling genetic algorithm of the present invention inverting obtains is consistent with the result that microhabitat serial genetic inverse obtains, both maximum differences are with respect to microhabitat serial genetic algorithm, consuming timely be about 56.37 hours for 100 generations of same model microhabitat master-slave mode paralleling genetic algorithm inverting, microhabitat serial genetic inverse is consuming time to be about 110.47 hours, it is the microhabitat serial genetic inverse twice that is roughly microhabitat master-slave mode paralleling genetic algorithm consuming time, this is because the fitness evaluation calculated amount is very big in the inverting of the BISQ of niche genetic algorithm model two-phase fissuted medium reservoir parameter, inverting is consuming time to be determined by the calculated amount of fitness evaluation basically, the microhabitat master-slave mode paralleling genetic algorithm of opening three processors has two to calculate fitness simultaneously from processor, and the microhabitat serial genetic algorithm of single processor has only a processor to calculate fitness.Calculate absolute speed-up ratio S according to the Amdahl law, be that the execution time of serial genetic algorithm is divided by the execution time of paralleling genetic algorithm, can get S=1.960, this has illustrated that microhabitat master-slave mode paralleling genetic algorithm has good concurrency, the time performance that has improved former niche genetic algorithm greatly (has only conveniently used less processor in order to say something herein, actual numerical value is more than processor number used herein far away when calculating, and counting yield can increase exponentially near linearity along with the increase of processor number at that time).

In sum, the present invention successfully is applied to microhabitat master-slave mode paralleling genetic algorithm in the finding the solution of BISQ model double phase anisotropic medium reservoir parameter inverse problem, and with respect to existing technology, its result of calculation and time efficiency all reach extraordinary effect.

The various embodiments described above only are used for explanation the present invention; wherein the structure of each parts, connected mode and manufacture craft etc. all can change to some extent; every equivalents and improvement of carrying out on the basis of technical solution of the present invention all should do not got rid of outside protection scope of the present invention.

Claims (8)

1. two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm may further comprise the steps:

1) according to calculation requirement, carries out building of MPI parallel computing platform; According to calculation requirement, MPICH is installed under the Windows/Linux system and disposes, carry out building of MPI parallel computing platform;

2) according to actual geologic information, make two-phase media BISQ model, as the implementation model of microhabitat master-slave mode paralleling genetic algorithm reservoir parameter inverting;

3) according to principle of least square structure objective function:

$E\left(p\right)=\frac{1}{2}{\left|\right|A\left(p\right)-G^{*}\left|\right|}^2,$

Wherein || || be the L-2 norm, p=[ρ _f, ρ _s, φ] ^TWait to ask parameter stream phase density ρ for discretize _f, the solid phase density p _sWith the one-dimensional vector that factor of porosity φ forms in a certain order, A is a vector valued function A:p → G, and G is the synthetic x component of theory of face of land medium solid phase displacement response and the one-dimensional vector that the z component forms according to a definite sequence, G ^*Be the actual measurement data x component and the one-dimensional vector of z component according to the order formation identical with G of face of land medium solid phase displacement response;

4) the main operational factor of setting genetic algorithm comprises population scale N, crossover probability P _c, the variation probability P _m, maximum genetic evolution algebraically GEN _Max, the minimum iteration error error during convergence and each inverted parameters span;

5) open MPI parallel computing platform MPICH2, the algorithm operational factor in the step 4) is loaded into, Num process opened by system, comprise a host process and Num-1 from process, initialization of population utilizes the floating-point encoding method to produce initial population by primary processor, and population scale is N;

6) by primary processor according to centralized dispatching dynamic task allocation strategy divide mating group to each from processor, carry out fitness according to the objective function of described step 3) structure and calculate, treat all from the processor fitness calculate finish after again by primary processor collection result;

7) use Sharing Function to adjust each individual fitness in the population, and in the population evolutionary process after this, the new fitness after algorithm is adjusted according to this is selected to eliminate computing with microhabitat;

8) fitness that at first step 7) calculated of primary processor carries out the conversion of sigma cut-off scales and obtains new fitness, according to this fitness population is not had the playback remainder then and selects computing at random, and adopt optimum conversation strategy, obtains body set one by one;

9) primary processor individual collections that step 8) is selected is done simulation scale-of-two crossing operation;

10) primary processor carries out the non-uniform mutation computing to the population that the step 9) crossing operation obtains, and generates progeny population;

11) step 10) is made a variation result that computing obtains carries out the reservoir parameter Inversion Calculation according to the objective function of step 3) structure, if meet end condition, then enter next step, continue to calculate as initial population otherwise the result that the computing that will make a variation obtains turns back to step 6);

12) will satisfy the result of calculation of end condition as the net result output of BISQ model double phase anisotropic medium reservoir parameter inverting, EOP (end of program).

2. the two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm as claimed in claim 1, it is characterized in that: the centralized dispatching dynamic task allocation strategy in the described step 6) refers to that system is divided into a primary processor and some from processor, the whole population of main processor monitors, calculation stages at fitness, by primary processor the dispensed of fitness is carried out from processor to each, when one when idle from processor, tell primary processor by message, primary processor will take out one by one body and distribute to this and proceed fitness from processor and calculate from the remaining individuality of population, fitness calculating up to whole population finishes, and collects the result by primary processor again.

3. the two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm as claimed in claim 1 or 2, it is characterized in that: in the described step 6), carry out fitness when calculating according to the objective function of described step 3) structure, at first to obtain the face of land medium solid phase displacement response of actual measurement as the inverting foundation, then at the parameter solid phase density p of wanting inverting _s, stream phase density ρ _fWith factor of porosity φ, two-phase media reservoir parameter inverse problem is considered as p (ρ in three dimensions _f, ρ _s, φ) seek 1 p ^*(ρ _f ^*, ρ _s ^*, φ ^*), make the theory of its corresponding face of land medium solid phase displacement response synthesize the face of land medium solid phase displacement response that under the least square meaning, is fit to actual measurement best, at last according to principle of least square structure objective function.

4. as claim 1 or 2 or 3 described two-phase media parameter inversion methods based on microhabitat master-slave mode paralleling genetic algorithm, it is characterized in that: in the described step 7), Sharing Function is a function of substantial connection degree between two individualities in the expression colony, with sh (d _Ij) expression, the fitness of the individuality after adjusting is that its former fitness value is divided by sharing degree m _i, namely

F'(X _i)＝F(X _i)/m _i

Wherein, F (X _i) be individual X _iFormer fitness value; m _iBe the Sharing Function sh (d between other individualities in individual i and the colony _Ij) sum, represent a kind of tolerance of individual i degree of share in colony; m _iExpression formula as follows:

$m_i=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{sh}\left(d_{\mathrm{ij}}\right)$ (i＝1,2,…,N)。

5. as each described two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm of claim 1～4, it is characterized in that: in the described step 8),

The conversion of described Sigma cut-off scales be with original fitness function F hint obliquely at for

Wherein c is a very little integer, and general value is c ∈ [1,5], and the value of c makes all fitness all more than or equal to 0,

With σ be respectively mean value and the variance of population fitness;

The specific operation process that described no playback remainder is selected at random is: at first calculate the existence desired number M of each individuality in colony of future generation in the colony _i,

, get M then _iIntegral part [M _i] be corresponding individual existence number in colony of future generation, can determine so altogether in the N of future generation colony

Individuality.At last with

Be the new fitness of each individuality, use and determine in the colony of future generation also undetermined at random than case selection method

Individuality;

Described optimum conversation strategy is that the highest fitness value of each colony of new generation that produces after for genetic manipulation and the highest fitness value of previous generation colony are made comparisons, if the highest adaptive value less than previous generation, just eliminate the body one by one in a new generation at random, the individuality of high adaptive value joins in a new generation having among the previous generation.

6. as each described two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm of claim 1～5, it is characterized in that: in the described step 9), the method for operating when doing simulation scale-of-two crossing operation is to the individual x of two fathers ₁And x ₂, generate two individual c of son in such a way ₁And c ₂:

$\left\{\begin{array}{c}{c}_{1,i}=[(1+\mathrm{\β})x_{1,i}+(1-\mathrm{\β})x_{2,i}]/2\\ c_{2,i}=[(1-\mathrm{\β})x_{1,i}+(1+\mathrm{\β})x_{2,i}]/2\end{array}\right.$ ，1≤i≤n

Wherein β is stochastic variable, all needs to regenerate on each dimension, and mode is as follows:

$\mathrm{\β}=\left\{\begin{array}{cc}{\left(2\mathrm{\μ}\right)}^{\frac{1}{\mathrm{\η}+1}}& u\≤0.5\\ {\left(2(1-\mathrm{\μ})\right)}^{-\frac{1}{\mathrm{\η}+1}}& u>0.5\end{array}\right.$

In the formula, μ is the random number that is uniformly distributed on the interval (0,1), and η is cross parameter, is an integer.

7. as each described two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm of claim 1～6, it is characterized in that: in the described step 10), generate in the process of progeny population, if carrying out by x=x ₁x ₂X _kX _nTo x'=x ₁x ₂X _k' ... x _nNon-uniform mutation when operation, if change point x _kThe genic value span at place is , new genic value x then _k' determined by following formula:

Wherein Δ (t, y) (y representative With

) expression meets a random number of non-uniform Distribution in [0, y] scope, requires the increase along with evolutionary generation t, (t y) also increases close to 0 probability Δ gradually.

8. as each described two-phase media parameter inversion method based on microhabitat master-slave mode paralleling genetic algorithm of claim 1～7, it is characterized in that: the end condition of described step 11) refers to that objective function satisfies trueness error error requirement or genetic evolution algebraically reaches maximal value GEN _Max

Images