CN104750657A - Numerical simulation redundancy parallel computing method applicable to fracture-cavity type structure carbonate reservoirs - Google Patents

Abstract

The invention belongs to the field of development of oil and gas fields and provides a numerical simulation redundancy parallel computing method applicable to fracture-cavity type structure carbonate reservoirs. The method includes: subjecting an established geologic model to domain decomposition to form n grid blocks; enabling each CPU (central processing unit) processor to be responsible for one grid block with one or more grids; after domain decomposition is finished, if adjacent grids exist at the domain decomposition boundary and a first grid is distributed to a first CPU processor while a second grid adjacent to the first grid is distributed to a second CPU processor, distributing the first grid to the second CPU processor and the second grid to the first CPU processor by means of redundancy.

Description

Be applicable to the numerical simulation redundancy parallel calculating method of fractured-cavernous carbonate reservoir

Technical field

The invention belongs to oil-gas field development field, be specifically related to a kind of numerical simulation redundancy parallel calculating method being applicable to fractured-cavernous carbonate reservoir.

Background technology

The nonuniformity of fractured-cavernous carbonate reservoir is comparatively strong, because numerical simulation adopts finite difference method to calculate pressure and saturation degree, must use the operation to Jacobi matrix and computing.In the process of dematrix, the physical property of adjacent mesh may differ comparatively large, causes speed of convergence slow, affects the counting yield of digital-to-analogue.For this problem, occurred the research of logarithm value simulation parallel algorithm both at home and abroad, compared typical method as follows: the mode adopting PC cluster at hardware aspect, framework multiple stage calculation server, the connected mode between server adopts Ethernet mode.In addition, a management server is also needed to coordinate each calculation server.In software, adopt the mode that territory is decomposed, the oil reservoir grid model participating in calculating is decomposed into several zonings, and distributes to each calculation server by management server and calculate, the result of calculation of each step out gathers later, then carries out next step calculating.The relative unit computing velocity of these class methods is enhanced, subject matter is that the communication holding time between each calculation server CPU is longer, especially, when computation host CPU quantity increases to a certain degree, the communication between CPU is even also long than the time calculated.

Summary of the invention

The object of the invention is to solve the difficult problem existed in above-mentioned prior art, a kind of numerical simulation redundancy parallel calculating method being applicable to fractured-cavernous carbonate reservoir is provided, realize the process of refinement of opposite joint hole type numerical reservoir simulation parallel algorithm, mainly solve the Communication between the CPU affecting overall performance, shorten the numerical simulation parallel computation time of such oil reservoir, improve overall calculation efficiency.

The present invention is achieved by the following technical solutions:

Be applicable to a numerical simulation redundancy parallel calculating method for fractured-cavernous carbonate reservoir, comprise:

Region Decomposition is carried out to the geologic model established, is broken down into n gridblock; A gridblock is responsible for by a CPU processor; Each gridblock comprises one or more grids;

After carrying out Region Decomposition, if have adjacent grid at the boundary of Region Decomposition, wherein first grid is assigned to first CPU processor, when second adjacent with it grid is assigned to second CPU processor, then respectively by first grid redundancy on second CPU processor, by second grid redundancy on first CPU processor.

Described method comprises further: required memory is assigned on all CPU processors participating in calculating.

Each CPU processor is responsible for calculating row corresponding with the gridblock that it is responsible in Jacobi matrix, and every a line represents a parameter needing to separate.

Compared with prior art, the invention has the beneficial effects as follows: the communication link of existing numerical value parallel algorithm is given up, replace CPU redundant computation function.Develop fracture-pore reservoir simulation theory and method, made numerical simulation study personnel can carry out digital-to-analogue calculating more efficiently.

Accompanying drawing explanation

Fig. 1, numerical simulation redundancy parallel algorithm stress and strain model procedure chart of the present invention.

Fig. 2, numerical simulation redundancy parallel algorithm grid check off fruit figure.

Step block diagram on each CPU of Fig. 3.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further detail:

The present invention proposes a kind of method that in solution parallel algorithm newly, CPU communication time is long on the basis of typical territory decomposition parallel algorithm.

1. grid division

Grid division, by grid (this grid refers to the grid that the geologic model that establishes divides out), according to METIS, (METIS is a kind of reseau-dividing algorithm, is the algorithm based on parallel multi-core processor.) carry out Region Decomposition, grid division principle keeps load balance and the traffic reduced between processor and complexity.Carry out Region Decomposition according to METIS specific as follows: in the grid division of numerical simulation parallel algorithm, whole digital-to-analogue grid regions (grid namely previously set up) is separated into many gridblocks.If have n CPU processor for numerical simulation parallel computation, these gridblocks or simulated domain need to resolve into n part (each gridblock comprises one or more grids), decompose automatically carry out or manually carry out according to physical condition based on used processor number.

2. distributed Memory Allocation

In order to reduce single cpu maximum memory demand, the method uses distributed memory to distribute, namely required memory being assigned on all CPU participating in calculating (is approximate mean allocation, the number that METIS algorithm is adjacent grid according to each grid is distributed, object be in order to balancing gridding between traffic, improve computing velocity).This scheme ensure that the opening (namely with more multi-CPU can the problem of Xie Yue great) of software.

The algorithm of numerical simulation generates Jacobi matrix by finite difference, and the formula of method of finite difference is as follows,

$-\underset{i}{\mathrm{\Σ}}\frac{{\∂R}_n^{\mathrm{\κ},t+1}}{{\∂x}_i}{|}_p(x_{i,p+1}-x_{i,p})=R_n^{\mathrm{\κ},t+1}\left(x_{i,p}\right)$

The integrated form of formula is as follows:

$\underset{{\mathrm{\Γ}}_n}{\∫}{F}^{\mathrm{\κ}}\·\mathrm{nd\Γ}=\underset{m}{\mathrm{\Σ}}{A}_{\mathrm{nm}}{F}_{\mathrm{nm}}$

This formula comes from material balance principle, namely flows into grid equal with remaining liq quality in grid with the difference of the quality flowing out grid liquid.

The solution of this equation needs the Jacobian matrix (Jacobi matrix is the matrix formed after carrying out difference to grid property, does not refer to actual grid) used in linear algebra principle.(Fig. 3 show be exactly the step block diagram each CPU utilizing above formula carry out numerical simulation)

Each CPU processor is responsible for calculating row corresponding with grid that it is responsible for (namely in n gridblock to should that gridblock of CPU) in Jacobi matrix, time two adjacent mesh calculate, (that is, two adjacent mesh blocks constitute whole matrix to corresponding whole matrix.When calculating, need in advance by corresponding row solution respective in matrix out, just can solve whole matrix.), wherein every a line represents a parameter needing to separate, and if oil saturation is matrix a line, water saturation is also a line, etc. In Jacobi matrix, the calculating of element is divided into two parts substantially: the calculating (not needing to communicate) of single grid and the calculating (needing to communicate) of all relevant links.

3. the communication of calculation server CPU

In numerical simulation calculation, due to grid has been carried out territory decomposition, and calculated by each separate CPU, there will be a problem, namely need respectively to exchange data between the originally adjacent grid of different CPU calculating.Therefore, in computation process, when each time step is complete, need to carry out exchanges data between the different CPU with original adjacent mesh, this exchanges data is called the communication process between calculation server CPU.

After tested, the time that this communication process consumes depends on 2 points: the mode of grid cutting when 1) territory is decomposed, and as far as possible efficient way reduces the adjacent mesh number being assigned to different calculation server CPU the most.2) increase of calculation server CPU quantity, self framework of this and parallel computation forms contradiction, namely increases CPU number and should shorten computation process, but the prolongation of the communication process brought along with the increase of CPU quantity can delay computation process again.

Second point in above 2 is the subject matter of restriction oil reservoir numerical simulation computing.The present invention adopts the calculating of grid Redundant binary number, does not namely adopt conventional territory isolation Distribution Calculation grid (first the present invention carries out METIS Region Decomposition, then carries out redundancy to the adjacent mesh in the region after decomposition).When territory is decomposed, (refer to the grid that two gridblock edges are adjacent one another are for there being adjacent grid.) when distributing to different CPU, the Redundant binary number mode adopting adjacent mesh to exchange processes, due to the adjacent problem that there are communications exchange data, therefore communication can affect greatly calculating.Here process at the net boundary place of Region Decomposition, in the region of the other side, carry out redundancy (namely adjacent grid being copied) by one of them grid, the matter of time of communication can be avoided like this.

Fig. 1 is numerical simulation redundancy parallel algorithm stress and strain model procedure chart.The concrete grid geometry of this figure logarithm value simulate redundant parallel algorithm divides and is described, and Fig. 2 is numerical simulation redundancy parallel algorithm grid check off fruit figure, and this figure illustrates the net result of redundancy grid, namely the final network participating in calculating.

In Fig. 12,4 and 5,7 is that adjacent area (refers to grid adjacent one another are in two gridblocks, the region of the redundancy of formation.) in grid, 2,4 and 5,7 respectively calculate terminate after, need the calculation result data in exchange two regions.Like this, two regions just create communication, thus expend certain communication time.Method shown in present employing Fig. 2, namely 2,4 sides increase by 5,7 two grid datas, like this, and can by 2 in one's respective area, the result and 5 after 4 calculating, the result after 7 calculating realizes mutual.

The feature of this method is:

1) communication process relatively has slowly been avoided: the communication process of Ethernet carries out with frame, although frame per second is generally in the level of Millisecond, frame per second is certain, and when communication number of times increases, efficiency must reduce.If do not carry out the communication between calculation server CPU, then arithmetic speed can be increased substantially.

2) problem of redundant data amount: the cost of not carrying out communication creates redundant data, thus the calculated amount of single cpu can be increased.Through measuring and calculating, in the most complicated 3D grid situation, total amount of data increases by 1/3rd at most.The data volume of 1/3rd for the impact of the counting yield of CPU far below the communication time between CPU.After tested, be the numerical simulation of 30,000 for grid number, the speed of redundancy parallel calculating method is approximately the 2-3 of common parallel algorithm doubly.

The numerical simulation redundancy parallel calculating method that the present invention proposes, is given up the communication link of existing numerical value parallel algorithm, is replaced CPU redundant computation function.Develop fracture-pore reservoir simulation theory and method, made numerical simulation study personnel can carry out digital-to-analogue calculating more efficiently.

Be applied in numerical reservoir simulation parallel computation, be particularly useful for the situation of lattice number huge (as ten million grid) and geologic model comparatively complicated (as fracture-cavity type carbonate), at present both at home and abroad without similar approach, therefore the application prospect of the method is better.

In order to verify the validity of this method, work out the Numerical Simulation Program of the method, and contrasted with existing Parallel Numerical program.

Test duration is-17 days on the 15th June in 2012, test environment: DELL PowerEdge blade server, 12 blades, 24 CPU, 96 cores.Test problem size: grid number 38546.The contrast of numerical simulation parallel algorithm is as shown in table 1.

Table 1

As can be seen from Table 1, common parallel Method for Numerical Simulation is little with redundancy parallel method difference computing time within 8 CPU, reason is this calculation server is 8 cores, therefore communication does not need directly to use PCI passage through Ethernet in mainboard, PCI communication channel speed is much larger than ethernet communication speed, and the advantage therefore after redundancy parallel method elimination communication impact is also not obvious.And after reaching 16 CPU, communication time becomes clearly the impact of efficiency, the computing velocity of redundancy approach of the present invention is all at more than 2.5 times of common parallel method.

The present invention relates to numerical reservoir simulation parallel calculating method, belong to oil-gas field development field.The present invention has carried out improvement in essence to existing numerical reservoir simulation parallel algorithm, make use of the performance of CPU to greatest extent, eliminates Ethernet in exchanges data, the inferior position of communication aspect.This method is specially adapted to the numerical simulation of number of grid huge (as ten million grid) and the situation of the problem of digital-to-analogue comparatively complicated (as crack solution cavity acid carbonate rock) own.These situations all can cause using common parallel method for numerical simulation to cause longer consequence consuming time.And after the contrast using digital-to-analogue redundancy parallel algorithm and classic method, correctness and the validity of this patent of invention method can be verified.

Technique scheme is one embodiment of the present invention, for those skilled in the art, on the basis that the invention discloses application process and principle, be easy to make various types of improvement or distortion, and the method be not limited only to described by the above-mentioned embodiment of the present invention, therefore previously described mode is just preferred, and does not have restrictive meaning.

Claims (3)

1. be applicable to the numerical simulation redundancy parallel calculating method of fractured-cavernous carbonate reservoir, it is characterized in that: described method comprises:

Region Decomposition is carried out to the geologic model established, is broken down into n gridblock; A gridblock is responsible for by a CPU processor; Each gridblock comprises one or more grids;

After carrying out Region Decomposition, if have adjacent grid at the boundary of Region Decomposition, wherein first grid is assigned to first CPU processor, when second adjacent with it grid is assigned to second CPU processor, then respectively by first grid redundancy on second CPU processor, by second grid redundancy on first CPU processor.

2. the numerical simulation redundancy parallel calculating method being applicable to fractured-cavernous carbonate reservoir according to claim 1, is characterized in that: described method comprises further: required memory is assigned on all CPU processors participating in calculating.

3. the numerical simulation redundancy parallel calculating method being applicable to fractured-cavernous carbonate reservoir according to claim 2, it is characterized in that: each CPU processor is responsible for calculating row corresponding with the gridblock that it is responsible in Jacobi matrix, every a line represents a parameter needing to separate.