CN-122021012-A - Geostatistical modeling method uncertainty quantization method based on mutual information entropy framework

CN122021012ACN 122021012 ACN122021012 ACN 122021012ACN-122021012-A

Abstract

The invention discloses a geostatistical modeling method uncertainty quantification method based on a mutual information entropy framework, which relates to the technical field of geostatistical modeling quality control and comprises the steps of collecting spatial sample data of an estimation domain, determining a target method, constructing a parameter matrix for each target method, modeling each target method on the estimation domain for multiple times according to the parameter matrix to generate a plurality of realization modeling results, dispersing each realization modeling result on a unified grid to obtain a realization modeling result set of each grid, counting a plurality of realization modeling result distributions for each grid, constructing probability distribution of each grid, calculating point entropy of each grid based on the probability distribution, and forming a spatial uncertainty entropy field. According to the invention, the quantification of the space uncertainty is realized by realizing the probability distribution and the entropy measurement, and the evaluation of the method output convergence and the structural consistency is realized by constructing the IMUI index by combining the entropy, the conditional entropy and the mutual information.

Inventors

HUANG YUHAN
SONG LI
YANG WEI
WANG JI
ZHAO YONGXUAN
LV QI
WU BIN
AO YINGCHUN
WANG MENGQI

Assignees

中国黄金集团数智科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A geostatistical modeling method uncertainty quantization method based on a mutual information entropy framework is characterized by comprising the steps of, Collecting space sample data of an estimation domain, determining target methods, constructing a parameter matrix for each target method, modeling each target method for a plurality of times on the estimation domain according to the parameter matrix, and generating a plurality of modeling results; Dispersing each realization modeling result on a unified grid to obtain a realization modeling result set of each grid, counting a plurality of realization modeling result distributions for each grid, constructing probability distribution of each grid, calculating point entropy of each grid based on the probability distribution, and forming a space uncertainty entropy field; Summing the point entropies of all grids to obtain the overall information entropy of the target method, selecting two sets of realization pairs from a plurality of realizations of the same target method as realization pairs, calculating the joint result distribution of the realization pairs, calculating the joint entropy according to the joint result distribution, calculating the conditional entropy of the realization pairs based on the joint entropy, and calculating the IMUI of the target method based on the overall information entropy and the conditional entropy; And calculating an IMUI (in-process model) for each target method, carrying out risk ranking on a plurality of target methods according to the IMUI, executing preprocessing operation on the space sample data when executing a risk reduction flow based on the IMUI, re-executing parameter matrix construction and re-calculating the IMUI of each target method based on the preprocessed space sample data, screening out high-risk target methods, and reserving low-risk target methods.
2. The method for quantifying the uncertainty of the geostatistical modeling method based on the mutual information entropy framework of claim 1, wherein the target method is each method of a set of geostatistical modeling methods and a set of interpolation methods to be evaluated; the set of geostatistical modeling methods to be evaluated comprises a geometric modeling method and a geostatistical modeling method.
3. The method for quantifying uncertainty of geostatistical modeling method based on mutual information entropy framework of claim 2, wherein said geostatistical modeling method is divided into a single solution type method and a multi-solution type method according to output form; The single solution method obtains a plurality of realizations through parameter disturbance, data resampling and random initialization, and the multiple solution method directly outputs the realizations.
4. The geostatistical modeling method uncertainty quantization method based on mutual information entropy framework of claim 3, wherein the unified grid is a regular grid construction method, and the estimation domain is divided into n×m uniform grids; counting the distribution condition of all the realization results for each grid to obtain a realization modeling result set; suppose at the first Individual grid cells Sharing in common The values of the realization results are respectively recorded as Order-making As an argument of the probability function, the probability function is constructed The expression of (2) is given as, ; Wherein, the Represent the first Personal grid The result of (2) is taken as Probability distribution at; Representing a mesh object; representing a grid index; representing the total number of implementation results; Index representing implementation result, take value 1 to 1 ; Representing variables The dirac function of (2) is used for calculating probability density distribution of specific point positions; Represent the first The first of the grids Obtaining a result value by realization; the point entropy of each grid is calculated based on the probability distribution, expressed as, ; Wherein, the Represent the first Personal grid Point entropy of (3); Is natural logarithm; summing the entropy of all grid points in the region to obtain the overall information entropy of the target method of the target region, wherein the expression is, ; Wherein, the Representing overall information entropy; two sets of realizations are optionally selected from the multiple realizations of the same target method as realization pairs, and a joint entropy is calculated, expressed as, ; Wherein, the Representation implementation And implementation Is a joint entropy of (2); Representation implementation And implementation Is a joint probability distribution of (1); And Representing the realization of the pair ) Implementation numbers of two groups of implementations; Conditional entropy is calculated according to the joint entropy, and the expression is that, ; Wherein, the Representation implementation And implementation Is a conditional entropy of (a).
5. The method for quantifying uncertainty of geostatistical modeling method based on mutual information entropy framework of claim 4, wherein said calculating IMUI of target method based on overall information entropy and conditional entropy is expressed as, ; Wherein, the Representing an index of uncertainty of the mutual information, The larger the target method is, the worse the suitability of the target method and the data is, and the higher the uncertainty is; Representing the sum of conditional entropy.
6. The geostatistical modeling method uncertainty quantization method based on the mutual information entropy framework of claim 5, wherein the set of interpolation methods comprises different interpolation methods; the different interpolation methods comprise a geometric class method, a variation function method, a kriging method, a probability class method, a random field method and a Bayesian class method; When mathematical structure differences exist in different interpolation methods, the IMUI calculation adopts classified approximation processing, wherein the geometric method is based on a result realized by space segmentation and local independent regression processing, the variation function method and the Kerling method are used for constructing IMUI related expression based on approximate normal assumption and covariance structure assumption, and the probability method, the random field method and the Bayesian method are used for constructing IMUI related expression based on the common influence of priori uncertainty and posterior uncertainty.
7. The method for quantifying the uncertainty of a geostatistical modeling method based on a mutual information entropy framework of claim 6, wherein said target method is capable of outputting only a single solution and approximating an IMUI with a correlation between the single solution and the estimated result when only one estimated result is provided in engineering.
8. The method for quantifying uncertainty of geostatistical modeling method based on mutual information entropy framework of claim 7, wherein when multiple target methods are needed to be compared, each method in the geostatistical modeling method set and the interpolation method set to be evaluated is respectively and repeatedly executed to construct a parameter matrix, model multiple times to generate multiple realizations and IMUI calculation, and an IMUI value set corresponding to the multiple methods is obtained.
9. The method for quantifying uncertainty of geostatistical modeling method based on mutual information entropy framework of claim 8, wherein risk ranking is performed on each target method in the set of geostatistical modeling methods and the set of interpolation methods to be evaluated according to IMUI value sets corresponding to multiple methods, high risk target methods are screened out, and low risk target methods are reserved.
10. The method for quantifying uncertainty of geostatistical modeling method based on mutual information entropy framework of claim 9, wherein parameter optimization is performed inside the low risk target method, and if any condition that the IMUI decrease amplitude is smaller than a preset threshold and the parameter optimization run reaches a maximum optimization run is satisfied, parameter optimization is terminated, and a final IMUI, a spatial uncertainty entropy field and a method risk ranking result of each low risk target method are output.

Description

Geostatistical modeling method uncertainty quantization method based on mutual information entropy framework Technical Field The invention relates to the technical field of geostatistical modeling quality control, in particular to a geostatistical modeling method uncertainty quantification method based on a mutual information entropy framework. Background The accuracy of mineral resource reserves estimation is not only affected by the quality of the sample data, but also is closely related to the suitability of the selected estimation method to the data characteristics. The existing estimation method has the following problems that systematic method deviation is caused by geometric modeling and implicit modeling methods based on distance function output, geostatistical methods such as Kerling and conditional simulation generate smooth effects due to stability and distribution hypothesis limitation, data driving methods are easy to be subjected to fitting and sensitive to data structures, uncertainty assessment, selection and parameter setting of the methods mainly depend on experience judgment and error verification at present, effective means capable of quantifying modeling methods and data adaptation risks are lacked, and risks of method deviation cannot be objectively assessed. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a geostatistical modeling method uncertainty quantification method based on a mutual information entropy framework, which solves the problem that the modeling method cannot be quantified to adapt to risks and methodological deviations. In order to solve the technical problems, the invention provides the following technical scheme: The invention provides a geostatistical modeling method uncertainty quantization method based on a mutual information entropy framework, which comprises the steps of collecting space sample data of an estimation domain and determining target methods, constructing a parameter matrix for each target method, modeling each target method for multiple times according to the parameter matrix on the estimation domain to generate multiple implementation modeling results, dispersing each implementation modeling result on a unified grid to obtain an implementation modeling result set of each grid, counting multiple implementation modeling result distribution of each grid, constructing probability distribution of each grid, calculating point entropy of each grid based on the probability distribution to form a space uncertainty entropy field, summing the point entropy of all grids to obtain overall information entropy of the target methods, selecting two sets of implementation pairs from multiple implementations of the same target method as implementation pairs, calculating joint result distribution of the implementation pairs, calculating the condition entropy of the implementation pairs based on the joint entropy, calculating an IMUI (matrix of the target methods) based on the overall information entropy and the condition entropy, calculating the IMUI, sequencing multiple target methods based on the IMUI, carrying out risk sorting on the multiple target methods based on the IMUI, carrying out the high risk reduction on the pre-sample data, carrying out the high risk matrix processing on the space sample data, and carrying out the high risk matrix processing on the target method. As a preferable scheme of the uncertainty quantization method of the geostatistical modeling method based on the mutual information entropy framework, the invention comprises the following steps: the target method refers to each method in a geostatistical modeling method set and an interpolation method set to be evaluated; the set of geostatistical modeling methods to be evaluated comprises a geometric modeling method and a geostatistical modeling method. As an optimal scheme of the uncertainty quantization method of the geostatistical modeling method based on the mutual information entropy framework, the geostatistical modeling method is divided into a single solution type method and a multi-solution type method according to an output form; The single solution method obtains a plurality of realizations through parameter disturbance, data resampling and random initialization, and the multiple solution method directly outputs the realizations. As a preferred scheme of the uncertainty quantization method of the geostatistical modeling method based on the mutual information entropy framework, the unified grid is a construction method of adopting a regular grid, and an estimation domain is divided into n multiplied by m uniform grids; counting the distribution condition of all the realization results for each grid to obtain a realization modeling result set; suppose at the first Individual grid cellsSharing in commonThe values of the realization results are respectively recorded asOrde