Method and device for predicting lithospheric major element composition based on multiple physical property parameters

Abstract

The method includes: obtaining geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crus

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of prediction of major elements of the lithosphere, and in particular, to a method and device for predicting a lithospheric major element composition based on multiple physical property parameters. BACKGROUND

[0002] Knowing the chemical compositions of the earth's crust and the lithospheric mantle not only can distinguish between lithologic compositions thereof, but also are crucial for the formation and evolution of the lithosphere. Knowing the chemical compositions of the earth's crust and the lithospheric mantle may help determine the distributions of minerals and mineral deposits, which is conducive to the exploration and development of mineral resources. The research on the chemical compositions of the earth's crust and the lithospheric mantle is conducive to evaluating potential risks of geological disasters such as earthquakes and volcanic eruptions, thereby providing early warning and preventive measures, etc. The chemical structure of the upper crust can be obtained by geological / drilling sampling and measurement of its major element composition. However, it is usually difficult to directly obtain the chemical structures of the deep crust and the lithospheric mantle.

[0003] Since geophysical prospecting can reflect the anomalous features of a physical field on the lithospheric scale and the structure construction of the deep earth, the geophysical prospecting method is still mainly employed to recognize the structural features of the deep earth. How to obtain the internal chemical structure of the lithosphere through the correlation of the physical properties of rock with major elements is a problem to be urgently solved.

[0004] Certain achievements have been made in the research on the relationship between the physical property parameters and the major element composition of deep rock. These achievements may be roughly classified into two categories.

[0005] One category is to estimate a lithospheric composition according to a seismic detection result by measuring the physical property parameters of different types of rock by experiments and directly associating the measured physical property parameters with corresponding major element composition testing results to establish an empirical formula of correlation. For example, Vp and SiO2 of rocks of amphibolite facies and granulite facies under high temperature and high pressure are measured by experiments. An empirical relationship of correlation between Vp and SiO2 of the middle-lower crust under the pressure of 600 MPa is established, and the average Vp of the middle- lower crusts of 8 continental earth's crusts is counted. The average SiO2 composition of the middle-lower crusts of the 8 continental earth's crusts is then obtained with the empirical relationship. With the Vp laboratory measurement results of igneous rock and metaigneous rock within a pressure range of 0.6-1.0 GPa, Vp is associated with SiO2 and MgO using multiple linear regression to establish a function of igneous rock and metaigneous rock compositions free of the garnet component with Vp, and then the composition of high Vp rock of the lower crust of the east coast of the United States is estimated.

[0006] The other category is to estimate a lithospheric composition with a seismic detection result by establishing a database of thermodynamic parameters and elastic parameters of different types of rocks / minerals and establishing a relationship between rock major elements and a wave velocity by a phase equilibrium simulation method. For example, a database of physical parameters of 60 minerals stable under high temperature and high pressure is established. Elastic properties and densities of the minerals are calculated with relationships of a temperature with an expansion rate, a density, a bulk modulus, and a shear modulus and based on the third-order finite strain theory, and longitudinal and transverse wave velocities in a rock, and a density and major elements of the rock are calculated according to composition proportions of the minerals. Whole-Rock Interpretative Seismic Toolbox for Ultramafic Lithologies (WISTFUL) establishes a database including 4485 ultra-basic whole-rock compositions; their mineral constituents, densities, and longitudinal and transverse wave velocities within pressure (P) and temperature (T) ranges (P=0.5-6 GPa, T=200-1600°C) are calculated using Gibbs free energy minimization program Perple_X, and mineral constituents, a density pressure, and a temperature are determined by searching for values closest to the seismic wave velocities in the model; and then the measured seismic wave velocities in the upper mantle are converted to a temperature, a density, and constituents. A function relationship of Vp with a major oxide content under an alpha-quartz stable field within the crustal range is established by assembling a continental crustal constituent model, and the SiO2 composition or the like of the cratonic crust of North China is calculated according to seismic profile data.

[0007] However, these methods use only a single physical property parameter for constrained calculation, and the accuracy of their calculation results needs to be improved. SUMMARY

[0008] In view of the above, the present application provides a method and device for predicting a lithospheric major element composition based on multiple physical property parameters to solve the problem of poor accuracy of calculation of a major element composition relationship of deep rock with a single physical property parameter in the prior art.

[0009] To achieve the above objective, the present application provides the following technical solutions:

[0010] In a first aspect, a method for predicting a lithospheric major element composition based on multiple physical property parameters includes:

[0011] step 1, obtaining geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in a research region;

[0012] step 2, calculating a density in the research region according to the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity;

[0013] step 3, calculating a lithospheric temperature structure in the research region with the terrestrial heat flow data and the global crust model data, and correcting the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity with the lithospheric temperature structure in the research region to obtain a corrected longitudinal wave velocity and a corrected transverse wave velocity;

[0014] step 4, calculating a plurality of physical property parameters having a good correlation with major elements according to the density, the corrected longitudinal wave velocity, and the corrected transverse wave velocity; and

[0015] step 5, inputting the calculated plurality of physical property parameters having the good correlation with the major elements to a pre-trained machine learning regression model under pressures for regression prediction to obtain a lithospheric major element composition in the research region.

[0016] Optionally, step 3 specifically includes:

[0017] step 301, calculating a one-dimensional steady-state heat conduction equation with the terrestrial heat flow data and the global crust model data;

[0018] step 302, calculating lithospheric temperature structures corresponding to heat flow measurement points according to the one-dimensional steady-state heat conduction equation;

[0019] step 303, eliminating outliers with a mantle heat flow below a first threshold and a moho surface above a second threshold from the lithospheric temperature structures corresponding to the heat flow measurement points;

[0020] step 304, obtaining the lithospheric temperature structure in the research region according to the lithospheric temperature structures corresponding to the heat flow measurement points with the outliers eliminated; and

[0021] step 305, correcting the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity with the lithospheric temperature structure in the research region to obtain the corrected longitudinal wave velocity and the corrected transverse wave velocity.

[0022] Optionally, in step 301, the one-dimensional steady-state heat conduction equation is obtained by a back-stripping method.

[0023] Optionally, in step 304, a Kriging interpolation method is adopted to obtain the lithospheric temperature structure in the research region according to the lithospheric temperature structures corresponding to the heat flow measurement points with the outliers eliminated.

[0024] Optionally, in step 305, a correction formula for the longitudinal wave velocity and the transverse wave velocity is as follows:

[0025] VP0=Vp + DVp(Tz-T0)

[0026] VSO = VS + Dys(Tz — To)

[0027] where 1% represents the corrected longitudinal wave velocity; Vso represents the corrected transverse wave velocity; Vp represents a calculated longitudinal wave velocity in the research region; Vs represents a calculated transverse wave velocity in the research region; DVp represents a longitudinal wave correction coefficient; Dys represents a transverse wave correction coefficient; Tz represents a temperature at a certain depth; and To is 20°C.

[0028] Optionally, in step 5, a training process of the machine learning regression model specifically includes:

[0029] step 501, obtaining rock physical property data of a rock sample in the research region, where the rock physical property data includes a longitudinal wave velocity in the rock, a transverse wave velocity in the rock, and a rock density;

[0030] step 502, fitting a relationship between the longitudinal wave velocity in the rock or the transverse wave velocity in the rock and a pressure to obtain a fitted model of a plurality of sets of longitudinal and transverse wave velocities in the rock changing with pressure;

[0031] step 503, selecting data with a goodness of fit greater than a third threshold from the fitted model of a plurality of sets of longitudinal and transverse wave velocities in the rock changing with pressure, and calculating longitudinal wave velocities in the rock and transverse wave velocities in the rock at different depths, and calculating rock physical property parameters according to the longitudinal wave velocities in the rock and the transverse wave velocities in the rock at different depths and the rock density;

[0032] step 504, performing regression analysis on the rock physical property parameters and major elements needing to be calculated in the research region by polynomial fitting to obtain regression parameters; and

[0033] step 505, selecting rock physical property parameters having a good correlation with the major elements needing to be calculated in the research region as a training dataset for training a 4 plurality of machine learning regression models, and performing evaluation with the regression parameters to obtain the machine learning regression model under different pressures.

[0034] Optionally, in step 502, the fitted model is as follows:

[0035] V(P) = V0+DP-B0 exp(—kP)

[0036] where V(P) represents a longitudinal or transverse wave velocity in the rock under a pressure of P; Vo represents a predicted velocity under zero pressure when no pore / crack is present; D represents a rate of a velocity changing with pressure in a linear elastic state; Bo represents an initial velocity difference caused by the presence of a pore / crack under zero pressure; and k represents an attenuation constant of a velocity decreasing in a nonlinear elastic state.

[0037] Optionally, in step 503, the third threshold is 0.95.

[0038] Optionally, in step 503, the rock physical property parameters include a wave velocity ratio, a longitudinal wave velocity, a transverse wave velocity, a density, a Poisson's ratio, a first Lame coefficient, and a second Lame coefficient.

[0039] In a second aspect, a device for predicting a lithospheric major element composition based on multiple physical property parameters includes:

[0040] a data obtaining module configured to obtain geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in a research region;

[0041] a density calculation module configured to calculate a density in the research region according to the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profde / slice data of the transverse wave velocity;

[0042] a correction module configured to calculate a lithospheric temperature structure in the research region with the terrestrial heat flow data and the global crust model data, and correct the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity with the lithospheric temperature structure in the research region to obtain a corrected longitudinal wave velocity and a corrected transverse wave velocity;

[0043] a physical property parameter calculation module configured to calculate a plurality of physical property parameters having a good correlation with major elements according to the density, the corrected longitudinal wave velocity, and the corrected transverse wave velocity; and

[0044] a lithospheric major element composition prediction module configured to input the calculated plurality of physical property parameters having the good correlation with the major elements to a pre-trained machine learning regression model under pressures for regression prediction to obtain a lithospheric major element composition in the research region.

[0045] Compared with the prior art, the present application has at least the following beneficial effects:

[0046] The present application provides a method and device for predicting a lithospheric major element composition based on multiple physical property parameters. The method includes: obtaining geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in a research region, and calculating a density in the research region; calculating a lithospheric temperature structure in the research region with the terrestrial heat flow data and the global crust model data, and correcting the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity with the lithospheric temperature structure in the research region to obtain a corrected longitudinal wave velocity and a corrected transverse wave velocity; calculating a plurality of physical property parameters having a good correlation with major elements according to the density, the corrected longitudinal wave velocity, and the corrected transverse wave velocity, and inputting the calculated plurality of physical property parameters having the good correlation with the major elements to a pre-trained machine learning regression model under pressures for regression prediction to obtain a lithospheric major element composition in the research region. In the present application, the plurality of physical property parameters having the good correlation with the major elements are input to the pretrained machine learning regression model under pressures for regression prediction to obtain the lithospheric major element composition in the research region. The regression prediction with multiple parameters can constrain the chemical composition of the lithosphere more accurately compared with regression prediction with a single physical property parameter.

[0047] When the machine learning regression model under pressure is trained, the relationships of the physical property parameters such as the longitudinal and transverse wave velocities in the rock, the wave velocity ratio, the Poisson's ratio, the first Lame coefficient (x density), and the second Lame coefficient (x density) with the rock major elements are compared, and several physical property parameters having the closest relationship with a certain rock major element are selected for training the machine learning regression model. The trained machine learning regression model can greatly improve the accuracy of a variable to be predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] To illustrate the prior art and the present application more intuitively, exemplary drawings are given below. It should be understood that specific shapes and constructions shown in the drawings usually should not be regarded as limiting conditions in implementing the present application. For example, those skilled in the art are able to easily make conventional adjustment or further optimization to addition / deletion / affiliation, specific shapes, positional relationships, connection manners, dimensional proportion relationships, etc., of some units (components) based on the technical concepts disclosed in the present application and the exemplary drawings.

[0049] FIG. 1 is a flowchart of a method for predicting a lithospheric major element composition based on multiple physical property parameters provided in Embodiment 1 of the present application;

[0050] FIG. 2 is a structural schematic diagram of a method for predicting a lithospheric major element composition based on multiple physical property parameters provided in Embodiment 1 of the present application;

[0051] FIG. 3 is a schematic diagram illustrating a relationship of a longitudinal wave velocity and a pressure for a representative rock type in Wendeng-Alxa Left Banner profile provided in Embodiment 1 of the present application;

[0052] FIG. 4 is a schematic diagram illustrating relationships of representative rock physical property parameters and SiO2 in Wendeng-Alxa Left Banner profile under a pressure of 600 MPa provided in Embodiment 1 of the present application;

[0053] FIG. 5 is a schematic diagram of a density structure in a deep seismic sounding profile of Wendeng-Alxa Left Banner provided in Embodiment 1 of the present application;

[0054] FIG. 6 is a schematic diagram illustrating thicknesses of the thermal lithosphere of North China Craton provided in Embodiment 1 of the present application;

[0055] FIG. 7 illustrates a lithospheric SiO2 composition in a deep seismic sounding profile of Wendeng-Alxa Left Banner provided in Embodiment 1 of the present application;

[0056] FIG. 8 illustrates relationships of rock physical property parameters of representative rocks and MgO in Jiaodong region under a pressure of 600 MPa provided in Embodiment 1 of the present application;

[0057] FIG. 9 illustrates MgO contents in slices at different depths in Jiaodong region provided in Embodiment 1 of the present application;

[0058] FIG. 10 illustrates a three-dimensional lithospheric MgO composition in Jiaodong region provided in Embodiment 1 of the present application;

[0059] FIG. 11 illustrates SiO2 contents in slices at different depths in Jiaodong region provided in Embodiment 1 of the present application; and

[0060] FIG. 12 illustrates a three-dimensional lithospheric SiO2 composition in Jiaodong region provided in Embodiment 1 of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The present application is described in further detail below by specific embodiments with reference to the accompanying drawings.

[0062] In the description of the present application, unless otherwise specified, "a plurality of" means two or more. The terms such as “first”, “second”, and “third” in the present application are intended to distinguish between objects referred to, rather than to have special meanings in terms of technical connotation (e.g., should not be construed as emphasizing a degree of importance or an order, etc.). The expressions such as “comprise”, “include”, and “have” also mean “be not limited to” (some units, components, materials, steps, etc.).

[0063] The words such as “upper”, “lower,” “left” “right”, and “middle” used in the present application are intended to indicate approximate relative position relationships in order to facilitate intuitive understanding with reference to the drawings, rather than to absolutely define the positional relationships in actual products.

[0064] Embodiment 1

[0065] Referring to FIG. 1 and FIG. 2, this embodiment provides a method for predicting a lithospheric major element composition based on multiple physical property parameters, including the following steps.

[0066] In step SI, geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in a research region are obtained.

[0067] Specifically, the terrestrial heat flow data is mainly composed of longitudes and latitudes and surface heat flow data. The terrestrial heat flow data, heat generation rates and thermal conductivities of crustal layers of tectonic units, and a global crust model are used for calculating a lithospheric temperature distribution. The global crust model used in this embodiment is CRUST 1.0 database.

[0068] In step S2, a density in the research region is calculated according to the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity.

[0069] Specifically, the geophysical profile / slice data of the longitudinal wave velocity refers to information about a longitudinal wave propagation velocity in a rock of the earth's interior obtained by the geophysical technique, i.e., the longitudinal wave velocity. The geophysical profile / slice data of the transverse wave velocity refers to information about a transverse wave propagation velocity in a rock of the earth's interior obtained by the geophysical technique, i.e., the transverse wave velocity.

[0070] In this step, a density structure needs to be calculated with a velocity structure by an empirical formula. Taking a North China Craton region as an example, the density of the North China Craton region may be calculated by the following empirical formula (1): ( 2.78 + 0.56(1^, - 6.0).........(Vp <6.0)

[0071] p = <3.07 + 0.29(1^, - 7.0).........(6.0 <Vp <7.5) (1) ( 3.22 + 0.20(1^, - 7.5).........(Vp >7.5)

[0072] where p represents the density, and Vp represents the longitudinal wave velocity.

[0073] In step S3, a lithospheric temperature structure in the research region is calculated with the terrestrial heat flow data and the global crust model data, and the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity are corrected with the lithospheric temperature structure in the research region to obtain a corrected longitudinal wave velocity and a corrected transverse wave velocity.

[0074] A specific process of this step is as follows.

[0075] In step S301, a one-dimensional steady-state heat conduction equation is calculated with the terrestrial heat flow data and the global crust model data.

[0076] Specifically, in this step, the one-dimensional steady-state heat conduction equation is solved by a back-stripping method using the terrestrial heat flow data and the crustal layered structure of the global crust model (CRUST1.0).

[0077] In step S302, lithospheric temperature structures corresponding to heat flow measurement points are calculated according to the one-dimensional steady-state heat conduction equation.

[0078] The one-dimensional steady-state heat conduction equation in this step is specifically as follows: d2T

[0079] K— + A = 0 (2)

[0080] where K represents a thermal conductivity, in units of W m’^K’1; T represents a temperature, in units of K; Z represents a depth, in units of km; A represents a heat generation rate, in units of pWm'3.

[0081] In step S303, outliers with a mantle heat flow below a first threshold and a moho surface above a second threshold are eliminated from the lithospheric temperature structures corresponding to the heat flow measurement points.

[0082] Specifically, the outliers with the mantle heat flow below 7 mW-m'2 and the moho surface above 1300°C are eliminated in this step.

[0083] In step S304, the lithospheric temperature structure in the research region is obtained according to the lithospheric temperature structures corresponding to the heat flow measurement points with the outliers eliminated.

[0084] Specifically, by using a Kriging interpolation method in this step, the lithospheric temperature structure in the research region can be obtained with the lithospheric temperature structures corresponding to the heat flow measurement points with the outliers eliminated.

[0085] In step S305, the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity are corrected with the lithospheric temperature structure in the research region to obtain the corrected longitudinal wave velocity and the corrected transverse wave velocity.

[0086] Due to a great temperature difference between different regions in the lithosphere, this embodiment uses a correction method for longitudinal and transverse wave velocity structures in the lithosphere. By linear fitting of wave velocities in rocks and temperatures and obtaining average values thereof, a correction coefficient of the longitudinal wave velocity vs temperature is determined to be Dvp=-5.544xl0'4 and a correction coefficient of the transverse wave velocity vs temperature to be Dy,=-3.316xl0'4. The correction coefficients are substituted into the following formulas (3) and (4) to correct the longitudinal wave velocity and the transverse wave velocity, respectively, to obtain the longitudinal wave velocity and the transverse wave velocity at 20°C, i.e., VpQ and Vs0:

[0087] VP0 =VP+ DVp(Tz - To) (3)

[0088] VSO = VS + Dys(Tz — To) (4)

[0089] where 1% represents the corrected longitudinal wave velocity; Vso represents the corrected transverse wave velocity; Vp represents a calculated longitudinal wave velocity in the research region; Vs represents a calculated transverse wave velocity in the research region; DVp represents a longitudinal wave correction coefficient; Dys represents a transverse wave correction coefficient; Tz represents a temperature at a certain depth; and To is 20°C.

[0090] In step S4, a plurality of physical property parameters having a good correlation with major elements are calculated according to the density, the corrected longitudinal wave velocity, and the corrected transverse wave velocity.

[0091] Specifically, the physical property parameters include a wave velocity ratio, a Poisson's ratio, a second Lame coefficient, a first Lame coefficient, a second Lame coefficient x density, and a first Lame coefficient x density. Specific calculation formulas of the physical property parameters are as follows.

[0092] 0 = Vp / Vs (5)

[0093] ¢7 = 0.5^-(^-1)-1] (6)

[0094] p = pV2 (7) Z = pVp — 2p (8) pp = p2V2 (9) pZ = p2V2 - 2p (10)

[0095] where 0 represents the wave velocity ratio; Vp represents the longitudinal wave velocity; Vs represents the transverse wave velocity; p represents the density; ^represents the Poisson's ratio; 2 represents the first Lame coefficient; p represents the second Lame coefficient; pX represents the first Lame coefficient x density; and pp represents the second Lame coefficient x density.

[0096] In step S5, the calculated plurality of physical property parameters having the good correlation with the major elements are input to a pre-trained machine learning regression model under pressures for regression prediction to obtain a lithospheric major element composition in the research region.

[0097] In this step, a training process of the machine learning regression model is specifically as follows:

[0098] In step S501, rock physical property data of a rock sample in the research region is obtained, where the rock physical property data includes a longitudinal wave velocity in the rock, a transverse wave velocity in the rock, and a rock density.

[0099] Specifically, the rock physical property data mainly includes the rock density (p), a porosity (¢), and a longitudinal wave velocity (Vp) and a transverse wave velocity (½) under a different pressure-temperature.

[0100] In step S502, a relationship between the longitudinal wave velocity in the rock or the transverse wave velocity in the rock and a pressure is fitted to obtain a fitted model of a plurality of sets of longitudinal and transverse wave velocities in the rock changing with pressure.

[0101] Specifically, at the model training stage, since the collected pressure and temperature ranges of the transverse wave velocity and longitudinal wave velocity data in rocks are not unified, the wave velocity data needs to be preprocessed, and variations of the wave velocity with pressure are fitted. The fitted model used in this embodiment can reflect a relationship between a pressure and a wave velocity, and the parameters have good geological significance. The fitted model can expand the available pressure range for physical property parameters, and is expressed using a unified formula, and can calculate any wave velocity value under a pressure of 0-3 GPa:

[0102] V(P) = V0+DP-B0 exp(—fcP) (11)

[0103] where P represents the pressure; V(P) is a longitudinal or transverse wave velocity in the rock under the pressure of P; Vo represents a predicted velocity under zero pressure when no pore / crack is present; D represents a rate of a velocity changing with pressure in a linear elastic state (without a porosity); Bo represents an initial velocity difference caused by the presence of a pore / crack under zero pressure; and k represents an attenuation constant of a velocity decreasing in a nonlinear elastic state.

[0104] In step S503, data with a goodness of fit greater than a third threshold is selected from the fitted model of a plurality of sets of longitudinal and transverse wave velocities in the rock changing with pressure, and longitudinal wave velocities in the rock and transverse wave velocities 11 in the rock at different depths are calculated, and rock physical property parameters are calculated according to the longitudinal wave velocities in the rock and the transverse wave velocities in the rock at different depths and the rock density.

[0105] Specifically, the data having the goodness of fit R2>0.95 is selected in this step, and the pressures at different depths are calculated using the fitted model (11). For example, the pressure at the depth of 5 km is about 130 MPa which is substituted into the formula (11) to calculate the values of Vp and Vs as a data array for the depth of 5 km. Similarly, the data arrays under different pressures can be obtained.

[0106] After the values of Vp and Vs are calculated, the physical property parameters such as the wave velocity ratio, the Poisson's ratio, the second Lame coefficient, the first Lame coefficient, the second Lame coefficient x density, and the first Lame coefficient x density can be calculated with the longitudinal wave velocity in the rock, the transverse wave velocity in the rock, and the density. Calculation formulas for the physical property parameters are as shown in formulas (5) to (10).

[0107] In step S504, regression analysis is performed on the rock physical property parameters and major elements needing to be calculated in the research region by polynomial fitting to obtain regression parameters.

[0108] Specifically, in this step, regression analysis is performed on the rock physical property parameters and the major elements needing to be calculated by polynomial fitting to obtain the regression parameters such as a determination coefficient and a root mean square error.

[0109] In step S505, rock physical property parameters having a good correlation with the major elements needing to be calculated in the research region are selected as a training dataset for training a plurality of machine learning regression models, and evaluation is performed with the regression parameters to obtain the machine learning regression model under different pressures.

[0110] At this point, the processing of the related data of the rock in this step is finished, and the desired machine learning regression model has been trained.

[0111] The method for predicting a lithospheric major element composition based on multiple physical property parameters provided in this embodiment is described in detail below by taking “a lithospheric SiO2 composition in a deep seismic sounding profile of Wendeng-Alxa Left Banner” and “lithospheric three-dimensional MgO and SiO2 compositions investigated with velocity slices of Jiaodong region” as examples.

[0112] The lithospheric SiO2 composition in the deep seismic sounding profile of Wendeng-Alxa Left Banner is predicted as follows.

[0113] Firstly, the machine learning regression model is trained.

[0114] In step 1, rock physical property data in a North China Craton region that the deep seismic sounding profile of Wendeng-Alxa Left Banner stretches across, and corresponding rock major element contents, terrestrial heat flow data, heat generation rates and thermal conductivities at different lithospheric depth levels of tectonic units, and the global crust model (CRUST1.0) are collected.

[0115] In step 2, velocities and pressures in rocks are fitted to obtain a fitted curve expression (i.e., the fitted model) of wave velocities in the rocks changing with pressure. FIG. 3 shows partial fitting results of fitting on different rocks, and it can be seen that the fitting results R2 of the velocities and pressures in different rocks are closes to 1 with good fitting effect, where the fitted curve (a) is gneiss; the fitted curve (b) is amphibolite; the fitted curve (c) is granulite; the fitted curve (d) is diorite; the fitted curve (e) is gabbro; and the fitted curve (f) is dunite.

[0116] A total of 526 pieces of data are fitted in this step, where there are a total of 460 pieces of data with R2>0.95, including 230 pieces of data of Vp and 230 pieces of data of Vs. The fitted model can well reflect the relationship of the pressure and the wave velocity.

[0117] In step 3, a pressure is substituted into the fitted model with R2>0.95 to calculate a data array within the desired pressure range. For example, for the data array under 5 km, an average pressure value of 130 Mpa is substituted into the formula (11) to obtain 230 pieces of Vp and Vs data, which will be used as a set of data for next step of calculation.

[0118] In step 4, the physical property parameters such as the wave velocity ratio, the Poisson's ratio, the second Lame coefficient, the first Lame coefficient, the second Lame coefficient x density, and the first Lame coefficient x density are calculated with the longitudinal wave velocity in a rock, the transverse wave velocity in a rock, and the density. Regression analysis is performed on the physical property parameters of rocks and SiO2 under the pressure of 600 MPa. It is found that Vp, p, 2, and p of the rock have good correlations with the SiO2 content, with R2>0.3, as specifically shown in FIG. 4. In FIG. 4, (a) shows a Vp-SiO2 correlation relationship; (b) shows a Vs-SiO2 correlation relationship; (c) shows a density-SiO2 correlation relationship; (d) shows a Poisson's ratio-SiO2 correlation relationship; (e) a second Lame coefficient-SiO2 correlation relationship; and (f) shows a first Lame coefficient-SiO2 correlation relationship.

[0119] In step 5, four rock physical properties Vp, p, 2, and p are used as a dataset to be trained for machine learning analysis. A plurality of algorithms are selected for comparison, and it is determined that a Gaussian process regression model with R2=0.82 is selected. Different Gaussian process regression models are trained with the datasets under different pressures. At this point, the training of the machine learning regression model is finished.

[0120] The lithospheric SiO2 composition in the deep seismic sounding profile of Wendeng-Alxa Left Banner is predicted using the trained machine learning regression model.

[0121] The deep seismic sounding profile of Wendeng-Alxa Left Banner is selected in this embodiment. This profile passes through the whole North China Craton, and the geophysical profile Vp data of North China Craton can be obtained, whereas Vs data is obtained by tomography. Next, the wave velocity data of the Wendeng-Alxa Left Banner profile needs to be calculated and temperature correction is performed such that a physical property parameter dataset needed for predicting the lithospheric SiO2 composition is calculated. Specific steps are as follows.

[0122] In step 1, geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in Wendeng-Alxa Left Banner are obtained.

[0123] In step 2, a density structure in the Wendeng-Alxa Left Banner profile is calculated using the empirical formula (1). The density structure is as shown in FIG. 5.

[0124] In step 3, to perform temperature correction on the profile, a temperature structure of North China Craton needs to be obtained. The calculation of the temperature structure requires the terrestrial heat flow data in the North China Craton region, and heat generation rates and thermal conductivities of strata of tectonic units, and the crustal layered structure of CRUST1.0 is extracted.

[0125] Specifically, the terrestrial heat flow data in the North China Craton region is mainly from Compilation of heat flow data in the continental area of China (4th edition), including a total of 743 pieces of data mainly composed of longitudes and latitudes and surface heat flow data. The crustal layered structure with the precision of l°xl° can be extracted from CRUST 1.0, including the thicknesses and the depths of the deposit layer, the upper crust, the middle crust, and the lower crust. The heat generation rates and thermal conductivities of the tectonic units including the sedimentary cover, the upper crust, the middle crust, the lower crust, and the lithospheric mantle are as shown in Table 1.

[0126] Table 1 Heat Generation Rates and Thermal Conductivities at Different Lithospheric Levels of Tectonic Units That Wendeng-Alxa Left Banner Profile Stretches Across

[0127] Tectonic Region Thermal conductivity (Wm'k1) Heat Generation Rate A (pW m J) Sedimenta ry Cover Uppe r Crust Middl e Crust Lowe r Crust Lithospher ic Mantle Sedimenta ry Cover Uppe r Crust Middl e Crust Lowe r Crust Lithospher ic Mantle Bohai Bay Basin 2.200 3.000 2.800 2.600 2.900 1.380 1.240 0.860 0.310 0.000 Ordos Basin 2.622 2.300 2.510 2.510 2.900 1.548 1.670 0.860 0.250 0.000 Other regions 2.200 3.000 2.800 2.600 2.900 1.260 1.260 0.960 0.650 0.000

[0128] Thermal lithospheric structures corresponding to heat flow measurement points are obtained by solving the one-dimensional steady-state heat conduction equation. Outliers with the moho temperature above 1300°C and abnormal data with the mantle heat flow below 7 mWm'2 are eliminated from calculation results, and a contour map of thermal structures of the terrestrial lithosphere of North China Craton is obtained by the Kriging interpolation method. FIG. 6 is a diagram illustrating thicknesses of the thermal lithosphere of North China Craton, i.e., depths at the temperature of 1300°C.

[0129] Finally, Vp and Vs are separately corrected with the calculated lithospheric temperature structure to obtain Vpo and Vso at the temperature of 20°C.

[0130] In step 4, the other two physical property parameters A and p needing to be used are calculated with Vpo, Vso, and p. Thus, the four types of data to be predicted of Vpo, p, K and p in the Wendeng-Alxa Left Banner profile have been obtained and the temperature correction is completed.

[0131] In step 5, since the calculated data is applicable to a pressure range of 0-3 Gpa, calculation and prediction can be carried out within a depth range of 0-100 km. Next, the pressures of the data to be predicted are grouped, and the trained Gaussian process regression models under pressures are called for prediction to obtain the lithospheric SiO2 composition in the Wendeng-Alxa Left Banner profile, as shown in FIG. 7.

[0132] Lithospheric three-dimensional MgO and SiO2 compositions investigated with velocity slices of Jiaodong region are predicted as follows.

[0133] Jiaodong is the third largest gold ore concentration area in the world. Investigating the lithospheric three-dimensional chemical compositions of Jiaodong is conducive to revealing its gold mineralization mechanism and mineralization law and guiding ore prediction. Since the establishment of the SiO2 composition has been discussion in the above example, this example mainly involves the discussion on the methods of calculating MgO and establishing a three-dimensional lithospheric chemical composition.

[0134] Firstly, the machine learning regression model is trained.

[0135] In step 1, rock physical property data of Jiaodong, and corresponding rock major element contents, terrestrial heat flow data, heat generation rates and thermal conductivities of strata of tectonic units, and the global crust model (CRUST 1.0) are collected. Since the above example encompasses a wide range and contains the Jiaodong research region, the data used is part of data in the above example. However, the geophysical velocity slice information is slightly different: with the data from January 1, 2000 to December 31, 2020 from some network stations in the Southeast of North China (114°E-123°E and 29°N-39°N), Shandong, Jiangsu, and Anhui, and by near earthquake tomography when Vp-Vs direct waves arrive, Vp and Vs tomographic data at depths of 0-40 km is obtained.

[0136] In step 2, velocities and pressures are fitted.

[0137] In step 3, a pressure is substituted into the formula with R2>0.95 to calculate a data array within the desired pressure range. Since the slice data is at intervals of a depth of 5 km, the average pressure data of 5-40 km at intervals of 5 km (Table 2) is calculated in this example, thereby obtaining a total of 8 sets of data under different pressures.

[0138] Table 2: Average Pressures Corresponding to Different Depths

[0139] Depth (km) Average Pressure (Mpa) Depth (km) Average Pressure (Mpa) 5 130.732 25 678.846 10 261.464 30 822.171 15 392.196 35 981.294 20 535.521 40 1114.366

[0140] In step 4, physical property parameters such as the wave velocity ratio, the Poisson's ratio, the second Lame coefficient, the first Lame coefficient, the second Lame coefficient x density, and the first Lame coefficient x density are calculated with the longitudinal wave velocity in a rock, the transverse wave velocity in the rock, and the density. Regression analysis is performed on the physical property parameters of rocks and MgO under the pressure of 600 MPa. It is found that other physical property parameters of the rock than the Poisson's ratio have good correlations with MgO, as shown in FIG. 8. In FIG. 8, (a) shows a Vp-MgO correlation relationship; (b) shows a Vs-MgO correlation relationship; (c) shows a density-MgO correlation relationship; (d) shows a Poisson's ratio-MgO correlation relationship; (e) a second Lame coefficient-MgO correlation relationship; and (f) shows a first Lame coefficient-MgO correlation relationship.

[0141] In step 5, seven rock physical properties Vp, Vs, p, z, p, pX, andpp are used as a dataset to be trained for machine learning analysis. A plurality of algorithms are selected for comparison, and it is determined that a Gaussian process regression model with R2=0.66 is selected. The tomographic data of Jiaodong is slice data at a depth interval of 5 km in nature, and therefore, the machine learning regression model needs to be trained at intervals of 5 km, and the average pressure data used is as shown in Table 2.

[0142] The lithospheric three-dimensional MgO composition investigated with velocity slices of Jiaodong region is predicted using the trained machine learning regression model:

[0143] In step 1, geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in Jiaodong region are obtained.

[0144] In step 2, density structure information is calculated using the empirical formula to obtain the density structure of Jiaodong region.

[0145] Since the lithospheric temperature structure within the range of North China Craton has 16 been calculated in the above example, it is not recalculated in this example.

[0146] In step 3, Vp and Vs are separately corrected with the lithospheric temperature structure obtained in the above example to obtain Vpo and Vso at the temperature of 20°C. Another physical property parameters needing to be used are calculated with Vpo, Vso, and p as a dataset to be predicted.

[0147] In step 4, the dataset to be predicted corresponding to each slice is substituted into the corresponding Gaussian process regression model under the average pressure for prediction to obtain the lithospheric MgO composition at every depth of 5 km in Jiaodong region, as shown in FIG. 9.

[0148] In step 5, the three-dimensional lithospheric MgO composition is obtained using three-dimensional linear interpolation in Matlab, as shown in FIG. 10.

[0149] In step 6, similarly, the lithospheric SiO2 composition at every depth of 5 km in Jiaodong region is calculated, as shown in FIG. 11.

[0150] In step 7, the three-dimensional lithospheric SiO2 composition is obtained using three-dimensional linear interpolation in Matlab, as shown in FIG. 12.

[0151] This embodiment provides a method for predicting a lithospheric major element composition based on multiple physical property parameters, where the plurality of physical property parameters having the good correlation with the major elements are input to the pretrained machine learning regression model under pressures for regression prediction to obtain the lithospheric major element composition in the research region. The regression prediction with multiple parameters can constrain the chemical composition of the lithosphere more accurately compared with regression prediction with a single physical property parameter. Moreover, regression prediction is performed with the measured data of a sample to analyze the chemical position of the lithosphere. Compared with a method of simulating a wave velocity in a rock and major elements, the method used in the present application is more in accordance with the actual geological conditions.

[0152] In this embodiment, when the machine learning regression model is learned, the relationships of physical property parameters and major elements are investigated, and it is found that using a single physical property parameter to calculate the correlations of the major elements is unsatisfactory, and the prediction accuracy of predicting the chemical composition of the lithosphere using the machine learning regression model with multiple parameters is significantly improved. The trained model has good portability and can be applied to every suitable region. Also, training data may be flexibly adjusted according to different regional features to obtain a major element prediction model suitable for local use.

[0153] Embodiment 2

[0154] This embodiment provides a device for predicting a lithospheric major element composition based on multiple physical property parameters, including:

[0155] a data obtaining module configured to obtain geophysical profile / slice data of a longitudinal wave velocity, geophysical profile / slice data of a transverse wave velocity, terrestrial heat flow data, and global crust model data in a research region;

[0156] a density calculation module configured to calculate a density in the research region according to the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profde / slice data of the transverse wave velocity;

[0157] a correction module configured to calculate a lithospheric temperature structure in the research region with the terrestrial heat flow data and the global crust model data, and correct the geophysical profile / slice data of the longitudinal wave velocity and the geophysical profile / slice data of the transverse wave velocity with the lithospheric temperature structure in the research region to obtain a corrected longitudinal wave velocity and a corrected transverse wave velocity;

[0158] a physical property parameter calculation module configured to calculate a plurality of physical property parameters having a good correlation with major elements according to the density, the corrected longitudinal wave velocity, and the corrected transverse wave velocity; and

[0159] a lithospheric major element composition prediction module configured to input the calculated plurality of physical property parameters having the good correlation with the major elements to a pre-trained machine learning regression model under pressures for regression prediction to obtain a lithospheric major element composition in the research region.

[0160] For specific implementation contents of the modules in the device for predicting a lithospheric major element composition based on multiple physical property parameters, reference may be made to the definitions on the above method for predicting a lithospheric major element composition based on multiple physical property parameters, which will not be described here redundantly.

[0161] The technical features of the foregoing embodiments can be arbitrarily combined (as long as there is no contradiction between the combinations of these technical features). For brevity of description, not all possible combinations of the technical features of the foregoing embodiments are described. However, the embodiments not written explicitly should be construed as falling within the scope described in the description.

Images