CN115564136A

CN115564136A - Geothermal history fitting and productivity prediction method

Info

Publication number: CN115564136A
Application number: CN202211321786.8A
Authority: CN
Inventors: 田小明
Original assignee: Individual
Current assignee: Individual
Priority date: 2022-10-27
Filing date: 2022-10-27
Publication date: 2023-01-03

Abstract

The invention discloses a geothermal history fitting and productivity prediction method, which is characterized in that a numerical agent model with lower degree of freedom is used for history fitting under the condition of geological data shortage, so that the problem that the numerical model can not be established under the condition of geological data shortage is solved, an efficient gradient calculation method is applied in the history fitting process, and optimized parameters are searched in a large-area parameter range, so that the time of gradient calculation involved in the traditional history fitting process is saved, and the calculation time of yield prediction of the model after the traditional history fitting is saved.

Description

Geothermal history fitting and productivity prediction method

Technical Field

The invention relates to a numerical simulation history fitting method in the field of geothermy, in particular to a method for fitting a geological data lack condition.

Background

With the development of various underground energy resources, the numerical simulation technology is gradually added to project requirements as an important technical means, and it is very important to establish a numerical model capable of reliably predicting the production capacity. History fitting is a very important and necessary step before capacity forecasting is performed using the established data model. Most of the existing history fitting methods consume a large amount of computing resources, are very time-consuming in computation, and in many cases, involve a large amount of manual intervention to adjust parameters. However, in most cases, the geological data cannot be sufficiently obtained for reasonable parameter adjustment, and particularly in the early stage of the development of a geothermal reservoir, the geological data is often very lacking.

At present, the traditional method is difficult to carry out thermal numerical simulation calculation under the condition of lacking geological data. Even if a numerical model is established, the search of optimization parameters in a large area range is almost impossible in the traditional history fitting and manual parameter adjusting process, because the large area of manual parameter adjustment means that the running process of the numerical model needs to be carried out for a plurality of times. In some cases where the number of grids of the numerical model is large, it often takes several tens of hours or even several days to perform a single model running, and it is needless to say that each parameter adjustment in the manual parameter adjustment process means that at least one model running is required. The computation time taken for running the model many times required in the history fitting process is very unwieldy. If a gradient descent method is used for history fitting, a large number of model runs are required to obtain the required numerical gradient.

Furthermore, even after history fitting using the above-described conventional method, the trained numerical model still cannot calculate the predicted yield very efficiently when performing yield prediction. This is because the numerical models used are still too complex, and these models sometimes even have redundant degrees of freedom, i.e. there are a number of mutually independent model parameters.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: the method comprises the steps of establishing a numerical model under the condition of lacking geological data, and solving the problems that a traditional history fitting mode cannot search optimized parameters in a large-area parameter range, gradient calculation involved in the traditional history fitting process consumes a large amount of computer resources and calculation time, and the model subjected to traditional history fitting still consumes a large amount of computer resources and calculation time when yield prediction is carried out.

The technical scheme adopted by the invention for solving the technical problems is as follows: a geothermal history fitting and capacity prediction method is characterized by comprising the following steps: the method comprises the following steps: establishing a discrete interwell affinity agent model, determining the boundary of the geothermal reservoir block, determining the outer boundary of a numerical agent model by using the boundary frame, and meshing the geothermal reservoir model after the boundary of the model is determined; step two: selecting conductivity parameters between grids, and the production index of the well is a digital model control parameter; step three: initializing control parameters and providing initial values for a history fitting model; step four: screening agent models, wherein the agent models which are too far away from the actual measurement historical production data of the geothermal field need to be screened before history fitting; step five: calculating the gradient in the history fitting process by using an adjoint gradient method; step six: predicting the future yield trend of the geothermal blocks by using the well matched agent model; step seven: a regularization term is introduced in the history fitted function.

Preferably, a necessary condition needs to be satisfied in the mesh generation process: any two wells cannot be located on the same grid. In the case that this condition is met, the entire geothermal block is discretely subdivided using as coarse a mesh as possible.

Preferably, the model control parameters include conductivity parameters of the model, well production index, density of the subterranean fluid, viscosity of the subterranean fluid.

Preferably, the conductivity parameter is initialized as a function of the permeability, the contact area, and the distance between the centroids of the two adjacent meshes. For different computing unit systems, the corresponding unit conversion coefficient needs to be multiplied.

Preferably, the upper and lower boundaries of the proxy model calculation result are the upper and lower boundaries of the future predicted yield of the geothermal block.

Preferably, the regularization term includes two parts: one part is the square term of the difference between the model parameters and the geological data; the other part is the correlation between different geological points extracted from geological data.

Preferably, the correlation relationship may be generally described by a geological variogram model.

Drawings

FIG. 1 is a CPU calculated time length comparison curve of gradient calculation in a geothermal history fitting and capacity prediction method and the gradient calculation in a common conventional method.

FIG. 2 is a flow chart of a method for geothermal history fitting and capacity prediction.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.

The specific implementation mode of the invention is as follows:

the method comprises the following steps: establishing a discrete interwell affinity proxy model, firstly determining the rough boundary of the geothermal reservoir block, determining the outer boundary of a numerical proxy model by using the boundary frame, and then performing mesh subdivision on the geothermal reservoir model after the boundary of the model is determined. A necessary condition is required to be met in the subdivision process: any two wells cannot be located in the same grid. In the case that this condition is met, the entire geothermal block is discretely subdivided using as coarse a mesh as possible. The model established by the method reflects the connection relation among different wells in the block and the connection relation between the wells and the grid of the block.

Step two: the selection of the model control parameters requires that it be clear which parameters are the control parameters of the model before the history fitting is performed, and the control parameters are continuously updated and optimized in the history fitting process. Several common control parameters that are selected as part of the history fitting process are: (1) Conductivity parameter between grids, (2) production index of the well. In addition, the boundary dimensions of the model, and physical properties of the subsurface fluid such as density, viscosity, etc. may also be selected as control parameters. Of course, if more control parameters are used, this also increases the complexity of the model and the time required for history fitting.

Step three: and initializing the control parameters of the model, wherein the initialization of the control parameters is the initial values of the model for history fitting. The selection of the initial values of the history fitting directly influences the final parameter optimization result. Since the non-linear problem is naturally characterized by multiple local extremum optimal solutions, the choice of different initial values will tend to bias the final optimization parameter to an extremum point closer to the initial value. The conductivity parameters are here initialized depending on the permeability of two adjacent meshes, the contact area, and the distance between the centroids of the meshes. For different calculation unit systems, the corresponding unit conversion coefficient can be multiplied. In addition, the initialization of the well's production index may be approximated by the Peacheman's formula. For an unstructured grid, the grid step size in the Peacheman's formula in the x and y directions can be set to the side length of an equivalent square with the same area as the unstructured grid. It should be noted that the initialization of the above control parameters does not require accurate values to be obtained, since these initial values will be continually updated during the history fitting process to eventually reach the error within the desired range.

Step four: in order to make the built model general, the initialization of model parameters is to perform random sampling initialization within a very large parameter range, such as permeability and porosity of the mesh required in the last parameter initialization process. Not all proxy models generated by random sampling are suitable for history fitting. Proxy models that are too far from the geodetic field measured historical production data need to be screened first before history fitting. The method comprises the following steps: if the relative error between the calculated result of the proxy model and the actual observed yield data is less than 60%, the proxy model is selected as the candidate proxy model, and the proxy models with the error larger than 60% are screened out. The advantage of this is that it can avoid spending most of the computation time on the history fitting process of some proxy models that do not meet the screening condition, thus greatly saving computation time. This figure for a relative error of 60% can be adjusted for a specific geothermal block and parameter settings.

Step five: compared with the traditional numerical gradient calculation, the gradient calculation in the history fitting process adopts a concomitant gradient method to calculate the gradient in the history fitting process. The calculation method of the adjoint gradient is firstly given by the optimal control theory, and the detailed analysis and derivation process can refer to the discussion about the adjoint gradient method in the classical optimal control theory. Compared with the conventional method, the method can obviously reduce the CPU calculation time. And as the model control parameters are gradually increased (namely the complexity of the model is increased), the efficiency improvement effect of the method is more obvious. As can be seen from the following figure, for the calculation of a single gradient, the calculation efficiency of the method is improved by at least two to three orders of magnitude compared with the conventional method. The time consumption of calculation in the process of screening a large number of agent models and then performing history fitting on the agent models is greatly reduced, and the calculation time is controlled within an acceptable range.

Step six: and (3) capacity prediction, namely screening a series of agent models through the calculation of the steps and well matching the screened agent models with actual observed production data. These well-matched proxy models can be used to predict future production yield trends for the geothermal block. The upper and lower boundaries of the series of proxy model calculations are the upper and lower boundaries of the future predicted yield of the geothermal block, and the average value thereof can be used as an important reference value for the future yield of the block.

Step seven: the optimization of the agent model, along with the gradual enrichment of geological exploration data of the landed heat reservoir and the further accumulation of production data, can be further improved and enhanced. To account for these already acquired geological data in the proxy model, regularization terms may be introduced in the history-fitted objective function. The regularization term includes two parts: one part is the square term of the difference between the model parameters and the geological data; the other part is the correlation between different geological points extracted from geological data. These correlations, which are typically described by a geological variogram model, are embodied in a regularization term, which is in fact the inverse of a covariance matrix of these parameters with respect to each other. Similarly, in order to take into account the measurement error of the actual observed yield data, a matrix related to the measurement error may be introduced between the model calculation result of the history fitting objective function and the squared term of the difference of the actual measured data, and usually the matrix is the inverse of the diagonal matrix formed by the variance of the measurement error.

Claims

1. A geothermal history fitting and capacity prediction method is characterized by comprising the following steps: the method comprises the following steps: establishing a discrete interwell affinity agent model, determining the boundary of the geothermal reservoir block, determining the outer boundary of a numerical agent model by using the boundary frame, and meshing the geothermal reservoir model after the boundary of the model is determined; step two: selecting conductivity parameters between grids, and the production index of the well is a digital model control parameter; step three: initializing control parameters and providing initial values for a history fitting model; step four: screening agent models, namely screening agent models which are too far away from the actual measurement historical production data of the geothermal field before history fitting; step five: calculating the gradient in the history fitting process by using an adjoint gradient method; step six: predicting the future yield trend of the geothermal block by using the well matched agent model; step seven: a regularization term is introduced in the history fitted function.

2. The geothermal history fitting and capacity forecasting method according to claim 1, wherein a necessary condition is satisfied in the mesh generation process: any two wells cannot be located on the same grid, and in the event that this condition is met, the entire geothermal block is discretely subdivided using a very coarse grid.

3. The method of claim 1, wherein the model control parameters comprise conductivity parameters of the model, well production index, density of the subsurface fluid, viscosity of the subsurface fluid.

4. The method of claim 1, wherein the conductivity parameter is initialized according to permeability, contact area and distance between centroids of two adjacent grids, and the conductivity parameter is multiplied by the corresponding unit transformation coefficient for different computing unit systems.

5. The method of claim 1, wherein the upper and lower boundaries of the proxy model calculation result are upper and lower boundaries of the future predicted yield of the geothermal block.

6. The method according to claim 1, wherein the regularization term comprises two components: one part is the square term of the difference between the model parameters and the geological data; the other part is the correlation between different geological points extracted from geological data.

7. The method of claim 6, wherein the correlation relationship is generally described by a geological degradation function model.