Lithosphere thermal structure inversion method based on physical constraint and event recognition

Abstract

The invention belongs to the technical field of geophysical exploration and basin analysis, and provides a lithosphere thermal structure inversion method based on physical constraint and event recognition, which comprises the following steps of: constructing physical constraint conditions including lithosphere rheological constraint, equilibrium compensation constraint and thermodynamic consistency constraint; the method comprises the following steps: establishing a layered lithosphere heat conduction model, solving a transient heat conduction equation by adopting a finite difference method, realizing automatic identification and parameter initialization of a construction event by utilizing a machine learning regression model, constructing a multi-objective loss function, performing global optimization by adopting a differential evolution algorithm, and finally evaluating parameter uncertainty by adopting a Monte Carlo method; the problems of strong multiplicity, lack of physical rationality, low calculation efficiency and the like of a traditional inversion method are solved, and efficient and reliable inversion of the lithosphere thermal structure parameters is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of geophysical exploration and basin analysis, and particularly relates to a lithospheric thermal structure inversion method based on physical constraints and event identification. BACKGROUND

[0002] Lithospheric thermal structure inversion is a core key technology in basin analysis, oil and gas reservoir potential evaluation, and regional tectonic evolution history reconstruction. The core goal is to quantitatively back-propagate key parameters such as lithospheric extension coefficient, rifting period, and thermal flow evolution experienced by the lithosphere on a geological time scale through present observed geological and geophysical data (such as stratigraphic temperature, thermal flow value, tectonic subsidence history, etc.), so as to reveal the formation mechanism and evolution dynamics of the basin.

[0003] At present, the traditional inversion method in this field mainly relies on a trial-and-error fitting strategy based on a global optimization algorithm (such as differential evolution algorithm, Monte Carlo simulation, genetic algorithm, etc.). That is, by constantly adjusting the input parameters (such as crust and mantle extension coefficients, event occurrence time and duration), running the forward model to simulate the temperature history and subsidence history, and minimizing the difference between the simulated data and the observed data to find the optimal solution.

[0004] However, such traditional methods have long faced the following technical bottlenecks in theoretical models and practical applications: 1. Strong multi-solution problem: Different extension histories may produce similar temperature and subsidence responses, the inversion results depend on the initial model selection, leading to multiple geological interpretations based on the same set of data, affecting the reliability of the conclusions.

[0005] 2. Weak physical mechanism constraints, difficult to guarantee geological rationality: Existing inversion methods lack embedding and constraints on the basic physical mechanism of lithospheric deformation. For example, the inversion results are prone to unreasonable solutions in geological sense, such as strong stretching of the crust and no change in the mantle, ignoring the coupling relationship between crust and mantle stretching in the pure shear model. At the same time, the forward model simplifies or ignores key physical processes such as sediment burial and thermal advection, further weakening the authenticity and predictive ability of the results.

[0006] 3. Low computational efficiency: High-precision transient heat conduction forward model has high computational cost, and global optimization algorithm based on population iteration needs to call the model thousands of times, resulting in extremely long inversion time (hours to days), making it difficult to quickly and batch apply the method in exploration production, and also difficult to carry out uncertainty analysis and multi-scenario comparison research.

[0007] 4. The intelligent event recognition capability is missing in the construction event: the existing method usually requires the user to specify the extension event period in advance, but the basin rift period is difficult to determine accurately in actual research. In addition, the inversion system lacks the ability to automatically identify and extract the event signal implied in the observation data, and the entire inversion process relies heavily on the prior geological knowledge of the researcher, making it difficult to discover unexpected tectonic events.

[0008] 5. Difficulty in multi-parameter collaborative inversion and coupling relationship processing: a complete extension event needs to be described by multiple parameters such as crustal extension coefficient, mantle extension coefficient, start time, duration, etc. The complex coupling and trade-off relationship between parameters makes the optimization process prone to local extremum and difficult to converge.

[0009] Therefore, it is urgent to develop a new inversion method that can adaptively identify tectonic events, deeply integrate physical mechanism constraints, and significantly improve the robustness of inversion efficiency and results, thereby providing more reliable technical support for basin dynamics research and oil and gas exploration decision-making. SUMMARY

[0010] To solve the above technical problems, the present application provides a lithospheric thermal structure inversion method based on physical constraints and event recognition to solve the problems in the prior art, and the technical scheme adopted by the present application is: A lithospheric thermal structure inversion method based on physical constraints and event recognition, comprising the following steps: Step S1, data preprocessing and physical constraint construction, input observation data and construct physical constraint conditions and parameter search space; Step S2, multi-stage transient heat conduction forward modeling, solving the transient heat conduction equation by finite difference method, and calculating the thermal subsidence based on the principle of isostasy; Step S3, intelligent event recognition and parameter initialization, using machine learning regression model to analyze the observation data; Step S4, adaptive inversion optimization under physical constraint conditions; Step S5, result verification and uncertainty evaluation.

[0011] Further, in step S1, the input observation data includes temperature evolution history, tectonic subsidence history, sedimentary burial history and surface temperature history; when constructing the physical constraint conditions, the lithospheric rheological constraint, isostatic compensation constraint and thermodynamic consistency constraint are included; The parameter search space includes: defining the parameter range of the crustal extension coefficient , the mantle extension coefficient , the event start time , and the duration .

[0012] Furthermore, step S2 includes: Establish a layered lithospheric heat conduction model, including sedimentary layers, crustal layers, lithospheric mantle, and asthenosphere; Solve the transient heat conduction equation using the finite difference method: ; in, For thermal diffusivity, Thermal conductivity, For density, For specific heat capacity, Distribution of radioactive heat generation rate; For temperature; For time; For depth; The symbol is for partial differentials; Based on the principle of equilibrium, thermal settlement is calculated using the following equilibrium settlement formula: ; in, For time Heat deposition at any given time; For time The change in mass per unit area of ​​the rock column from the surface to the equilibrium compensation depth z at any given moment; , indicating time ,depth Instantaneous rock density at the location; This represents the initial density distribution; The density of the asthenosphere; Density of the filling material; The transient heat conduction equation is used to calculate the evolution of the temperature field under given tension event parameters; Density of temperature field corrected by thermal expansion ρ (T) is transformed into a density field change; The density field change was used to calculate the surface subsidence history using the equilibrium subsidence formula. .

[0013] Furthermore, step S3 includes: The observed data were analyzed using a machine learning regression model: ; in, The input feature vector includes an interpolation sequence of temperature evolution history and tectonic subsidence history; The output parameter vector includes the crustal stretching coefficient. Mantle tensile coefficient Event start time Duration and event validity identifier ; generating a training dataset, comprising: random sampling of event parameters: generating a plurality of stages of extension event parameters in a preset geological range, including event stage, start time of each event stage , duration , crustal extension coefficient and mantle extension coefficient ; forward modeling to generate response: inputting the plurality of stages of extension event parameters into step S2 to calculate the corresponding temperature evolution history and tectonic subsidence history ; adding observation noise: adding random noise conforming to normal distribution on the generated temperature evolution history and tectonic subsidence history ; constructing a training sample: taking the noise-added temperature evolution history and tectonic subsidence history sequence as input features , taking the original event parameter group of the data as output labels , to jointly constitute a training sample ; training a multi-output regression model to obtain a mapping function , establishing a mapping relationship from observation data to inversion parameters; generate the objective function as: ; wherein, is the mathematical expectation operator; is the true event parameter vector known in the synthetic data; is the predicted output of the machine learning model for the synthetic input features ; is the square of the norm; output the preliminary intelligent estimation values of the number of events, extension coefficient and time parameters.

[0014] Further, step S4 comprises: based on the intelligent estimation values of step S3, performing global optimization in the parameter search space of step S1 using a differential evolution algorithm, comprising: taking the temperature evolution history and tectonic subsidence history input in step S1 as the fitting target, constructing a multi-objective loss function: ; wherein: temperature fitting error: ; subsidence fitting error: ; Coupling penalty term: ; in, The value of the multi-objective loss function; These are the weighting coefficients; This is the penalty coefficient; This represents the temperature fitting error in the temperature history. This represents the number of temperature observation data points. For the first One observed temperature value; For the first One simulated temperature value; The settlement fitting error in the settlement history; This represents the number of settlement observation data points; For the first One observed settlement value; For the first One simulated settlement value; To calculate the crustal extension coefficient for all identified tectonic events. With mantle stretching coefficient The sum of squares of the differences.

[0015] Furthermore, step S5 includes: Perform a physical consistency check on the inversion results: ; in, The overall score is the physical consistency score; the higher the score, the better the inversion result conforms to the preset physical constraints. For the first The weight of each physical constraint; For the first Scoring of the degree to which physical constraints are satisfied; The Monte Carlo method is used to assess parameter uncertainty: ; in, is the posterior standard deviation of the inversion parameters; The number of valid sample groups obtained from Monte Carlo sampling; For the first The value of this parameter in the Monte Carlo sample group; For this parameter in all The mean of the group sample; Generate multiple sets of equivalent solutions, analyze the degree of ambiguity of the inversion results, and output the final inversion parameters and their confidence intervals.

[0016] The present invention has the following beneficial effects: (1) Intelligent event recognition capability: Automatically identify the period and time of tectonic events through machine learning technology, reducing reliance on prior geological knowledge; (2) Physical constraint guarantee: The physical mechanism of lithospheric deformation is introduced as a constraint condition to ensure the geological rationality of the inversion results; (3) Improved computational efficiency: Machine learning-assisted fast estimation provides a high-quality initial solution for global optimization, significantly reducing the number of iterations required for convergence; (4) Mitigation of multiple solutions: By reducing the dimension of the solution space through physical constraints and intelligent recognition, the generation of incompatible solutions is effectively suppressed; (5) Adaptive inversion capability: Automatically adjust the inversion strategy according to the data quality and type to achieve coordinated constraints on temperature and settlement data. Attached Figure Description

[0017] Figure 1 This is a flowchart. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0019] like Figure 1 A method for inverting the thermal structure of the lithosphere based on physical constraints and event recognition includes the following steps: Step S1: Data preprocessing and physical constraint construction, including: Input observational data: including temperature evolution history, tectonic subsidence history, sedimentary burial history, and surface temperature history; The physical constraints are constructed, including lithospheric rheological constraints, isostatic compensation constraints, and thermodynamic consistency constraints. The lithospheric rheological constraints are used to ensure that the lithospheric strength corresponding to the inversion parameters is mechanically reasonable, specifically including the upper limit constraint of the crustal stretching factor. ≤3.0), constraints on the coupling relationship between mantle stretching factor and crustal stretching factor ( ≥ (preferring a pure shear deformation mechanism), and the lower limit constraint of crustal thickness after stretching ( (>15km); The equilibrium compensation constraint requires that the inverted thermal structure can produce a settlement consistent with the observation through the equilibrium principle. Specifically, it includes the mass column conservation constraint before and after stretching, the consistency constraint between theoretical tectonic settlement and actual settlement, and the matching constraint between sedimentary load and settlement response. The thermodynamic consistency constraint mainly includes the surface heat flow range constraint (30-120mW / m²), the temperature range constraint of the lithosphere floor (1300±100℃), the rationality constraint of crustal temperature gradient (15-35℃ / km), and the monotonically increasing temperature field constraint, to ensure that the inversion results conform to the physical laws of the heat conduction equation and the thermal structure characteristics of a typical continental lithosphere.

[0020] Setting the parameter search space: Defining the crustal tensile coefficient Mantle tensile coefficient Event start time Duration The parameter range, this search space defines the parameter optimization boundary for subsequent global optimization algorithms.

[0021] Step S2: Multi-stage transient heat conduction forward modeling, including: A layered lithospheric heat conduction model was established, including sedimentary layers, crustal layers, lithospheric mantle, and asthenosphere; Solving the transient heat conduction equation using the finite difference method (assuming thermal conductivity) (constant) ; in, Thermal diffusivity, in km. 2 / Ma, Thermal conductivity, in W / (m•K). Density, unit: kg / m³ 3 ; Specific heat capacity, unit: J / (kg•K); Temperature, in °C or K; Time, in Ma; Depth, in km; This represents the distribution of radioactive heat generation rate, in μW / m³. 3 , The surface radioactive heat generation rate, in μW / m² 3 , D The radioactive heat generation rate decay depth is expressed in km.

[0022] A Lagrange depth tracking method is introduced to handle the material advection effect during the stretching process; This transient heat conduction equation is the core governing equation for lithospheric thermal evolution simulation. Its purpose is to calculate the evolution of the temperature field within the lithosphere over time and depth. This is achieved by providing an initial temperature field, boundary conditions (surface temperature history and lithosphere floor temperature), and extensional event parameters (…). Solving this equation yields the simulated temperature history. This refers to the temperature distribution at different depths within the lithosphere during different geological periods. This temperature field forms the basis for subsequent calculations of thermal expansion, density changes, and isostatic settlement.

[0023] Thermal settlement is calculated based on the principle of equilibrium. The formula for calculating equilibrium settlement is: ; Formula parameter description: For time The amount of heat settlement at any given time, expressed in meters (m). For time The change in mass per unit area of ​​the rock column from the surface to the equilibrium compensation depth z at any given moment; , indicating time ,depth The instantaneous rock density at a given location, expressed in kg / m³; this density is a function of temperature, expressed by the thermal expansion equation. Calculated, where The solution is obtained from the transient heat conduction model. For reference temperature, the unit is the same. , Reference temperature The density of the rock below is expressed in kg / m³. The coefficient of thermal expansion is K. -1 ; To ensure balanced compensation depth, the unit is meters (m). The initial density distribution specifically refers to the moment after the tectonic stretching event ends and the thermal relaxation process begins. The density of the lithosphere varies with depth. The distribution is expressed in kg / m³, which serves as the baseline reference state for settlement calculation. This represents the density of the asthenosphere, expressed in kg / m³. Density of the filling material, in kg / m³; The purpose of this equilibrium subsidence formula is to transform the evolution of the temperature field into an observable history of surface tectonic subsidence. Its calculation principle is: temperature changes cause thermal expansion or contraction of rocks, thereby altering the rock density (formula). Changes in density lead to variations in the mass per unit area of ​​the rock column, resulting in surface subsidence or uplift under an isostatic compensation mechanism. The calculation results of this formula represent a simulated subsidence history. This refers to the tectonic subsidence of the basin basement over time. This subsidence history can be compared with observed tectonic subsidence history and is one of the key observational data for constraining inversion parameters.

[0024] The connection between the two formulas and model construction: The two formulas mentioned above together constitute a complete thermodynamic coupled forward model. The transient heat conduction equation simulates the thermal process, and the equilibrium sedimentation formula simulates the mechanical response. The two are closely coupled through the temperature-density relationship (thermal expansion effect). Specifically: The transient heat conduction equation calculates the temperature field evolution under given tension event parameters. ; Density formula for temperature field corrected by thermal expansion This is transformed into a change in density field; The density field change was used to calculate the surface subsidence history using the equilibrium subsidence formula. .

[0025] This coupling process ensures that the model simultaneously satisfies the physical laws of heat conduction and the principle of equilibrium compensation, giving the simulation results (temperature history and sedimentation history) clear physical meaning and providing a solid physical foundation for subsequent inversion.

[0026] Step S3: Intelligent event recognition and parameter initialization, including: Machine learning regression models are used to quickly analyze the observed data: ; in, The input feature vector contains an interpolation sequence of temperature evolution history and tectonic subsidence history; The output parameter vector contains (Crustal tensile coefficient) (Mantle stretching coefficient) (Event start time) (Duration) and (Event validity identifier) ​​and other parameters.

[0027] The strategy for generating the training dataset is as follows: 1. Random sampling of event parameters: Within a preset reasonable geological range, multiple phases of extensional event parameters are randomly generated, including the event phase (phases 1-4) and the start time of each event. Duration crustal tensile coefficient With mantle stretching coefficient .

[0028] 2. Forward modeling to generate response: Input the parameters of multiple tensile events into the forward model described in step S2 to calculate the corresponding temperature evolution history. and tectonic subsidence history .

[0029] 3. Add observation noise: To simulate the uncertainty of real observation data, in the generated... and Random noise that follows a normal distribution is superimposed on top.

[0030] 4. Construct training samples: Use the noisy temperature evolution history and tectonic subsidence history sequences as input features. The original event parameter set used to generate this data will be used as the output label. Together they constitute a training sample By repeating the above process tens of thousands to hundreds of thousands of times, a large-scale synthetic training dataset covering a wide range of geological scenarios was constructed.

[0031] 5. Train the multi-output regression model to obtain the mapping function. In a specific form, a rapid mapping relationship is established from observation data to inversion parameters.

[0032] The objective function for generating synthetic data is: ; Formula parameter description: is the mathematical expectation operator, which represents averaging over all samples in the synthetic training dataset. The vector of known real event parameters (labels) in the synthetic data. For machine learning models to synthesize input features The predicted output. for The square of the norm is the sum of the squares of the errors.

[0033] The system outputs preliminary intelligent estimates of the number of events, the tension coefficient, and time parameters; the number of events is determined by the model output. The number of parameter groups is determined, and the tension coefficient and time parameter define the spatial and temporal attributes of each recognition event; Event effectiveness evaluation mechanism: When the event parameters predicted by the machine learning model meet the following conditions: (Significant tension event); Ma (duration in a geological sense); Prediction confidence (Model output probability); The parameter combination conforms to prior geological knowledge (such as...) ); When the predicted event parameters do not meet the above conditions, or are identified as false signals caused by data noise / measurement error; In subsequent optimization steps, only for Physical constraint optimization is performed on valid events. Such events are excluded to avoid interference from the spurious solution space; The intelligent estimate output in this step will serve as a high-quality initial solution for subsequent global optimization. Step S4: Adaptive inversion optimization under physical constraints, including: Based on the intelligent estimate obtained in step S3, global optimization is performed within the parameter search space of step S1 using a differential evolution algorithm, including: Using the temperature evolution history and tectonic subsidence history input in step S1 as fitting targets, a multi-objective loss function is constructed. This function simultaneously measures the fitting error between the simulation results and the observed temperature evolution history and the observed tectonic subsidence history. The multi-objective loss function is: ; in, Temperature fitting error: ; Settlement fitting error: ; Coupling penalty term: ; in, It is a multi-objective loss function, which is the objective of optimization (minimization). , These are the weighting coefficients. This is the penalty coefficient. The baseline weights for temperature fitting ensure that the basic constraints of the thermal structure are met. The weighting of subsidence fitting is higher than that of temperature, reflecting the importance of tectonic subsidence data for basin analysis (subsidence data usually has higher measurement accuracy and geological significance). Sensitivity analysis revealed that this range can effectively constrain the consistency of crust-mantle deformation while avoiding excessive punishment of reasonable deformation differences. This represents the temperature fitting error, specifically the root mean square error of the temperature history. This represents the number of temperature observation data points. For the first One observed temperature value. For the first A simulated temperature value. This represents the settlement fitting error of the settlement history, i.e., the root mean square error of the settlement history. This represents the number of settlement observation data points. For the first One observed settlement value. For the first One simulated settlement value. For all identified construction events (subscripts) Calculate its crustal tensile coefficient. With mantle stretching coefficient The sum of squares of the differences is used to penalize geologically unreasonable strong decoupling events.

[0034] During the optimization process, each evaluation of the loss function requires calling the forward model in step S2 to calculate the simulated temperature history and sedimentation history based on the current parameters. The differential evolution algorithm continuously updates the parameter population within the parameter search space to minimize the loss function, thereby obtaining the optimal inversion parameters.

[0035] A differential evolution algorithm is used for global optimization, incorporating physical constraints as boundary conditions; Implement adaptive parameter adjustments and dynamically adjust the search strategy according to the inversion progress; Step S5: Result Validation and Uncertainty Assessment Perform a physical consistency check on the inversion results: ; in, Physical consistency score: The higher the score, the better the inversion result conforms to the preset physical constraints. , , These are the weights of the lithospheric rheological constraints, the equilibrium compensation constraints, and the thermodynamic consistency constraints, respectively. , , Each constraint satisfaction level is scored. The specific calculation method is as follows: Lithospheric rheological constraint score ( ): ; in: (Reasonableness score of crustal extension coefficient); (Score of mantle-crust coupling relationship); (Score for the rationality of crustal thickness after stretching). Equilibrium compensation constraint score ( ): ; in: (Mass conservation score) This is the integral result of the change in mass per unit area of ​​the rock column, in kg / m². , represents the initial mass of the rock column, in kg / m²; (Settlement consistency score); (Load-settlement matching score); Thermodynamic consistency constraint score ( ): ; in: (Surface heat flow rationality score) (Reasonableness score for temperature at the base of the lithosphere); (Temperature gradient rationality score) Weighting: (Rheological constraint weights); (Balance constraint weights); (Thermodynamic constraint weights); in, The value ranges from 0 to 1, where 1.0 indicates that it fully complies with all physical constraints, and 0.0 indicates that it seriously violates the laws of physics.

[0036] The Monte Carlo method is used to assess parameter uncertainty: ; Formula parameter description: For a certain inversion parameter (e.g.) The posterior standard deviation of is used to measure its uncertainty. The number of valid sample groups obtained from Monte Carlo sampling. : No. The value of this parameter in the Monte Carlo sample group. This parameter is in all The average value of the sample group. Generate multiple equivalent solutions, analyze the degree of ambiguity of the inversion results; output the final inversion parameters and their confidence intervals.

[0037] Preferably, the machine learning regression model described in step S3 uses random forest, gradient boosting, or neural network algorithms to establish a nonlinear mapping relationship from temperature-sedimentation sequence to tension parameters through synthetic data training.

[0038] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, alterations, alterations, or substitutions made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A lithospheric thermal structure inversion method based on physical constraints and event recognition, characterized in that, The method comprises the following steps: Step S1, data preprocessing and physical constraint construction, input observation data and construct physical constraint conditions and parameter search space; Step S2, multi-stage transient heat conduction forward modeling, finite difference method is used to solve transient heat conduction equation, and heat subsidence is calculated based on the principle of isostasy; Step S3, intelligent event identification and parameter initialization, machine learning regression model is used to analyze observation data; Step S4, adaptive inversion optimization under physical constraint conditions; Step S5, result verification and uncertainty evaluation.

2. The lithospheric thermal structure inversion method based on physical constraints and event recognition according to claim 1, characterized in that, In step S1, the input observation data includes temperature evolution history, tectonic subsidence history, sedimentary burial history and surface temperature history; when constructing physical constraint conditions, lithospheric rheology constraint, isostatic compensation constraint and thermodynamic consistency constraint are included; The parameter search space includes parameter ranges defining a crustal extension coefficient , a mantle extension coefficient , an event start time , a duration .

3. The lithospheric thermal structure inversion method based on physical constraints and event recognition according to claim 1, characterized in that, Step S2 includes: A layered lithospheric heat conduction model is established, including sedimentary layer, crust layer, lithospheric mantle and soft flow layer; Finite difference method is used to solve transient heat conduction equation: ； wherein, is the thermal diffusivity, is the thermal conductivity, is the density, is the specific heat capacity, is the radioactive heat generation rate distribution; is the temperature; is the time; is the depth; is the partial differential symbol; Based on the principle of isostasy, heat subsidence is calculated by the following isostatic subsidence formula: ； where is time the thermal subsidence at time is time the change in mass per unit area of the column of rock from the surface to the isostatic compensation depth z at time , represents the instantaneous rock density at time , depth is the initial density distribution is the density of the soft mantle is the density of the fill​ Transient heat conduction equation calculates the evolution of temperature field under given extension event parameters; Temperature field is converted to density field changes by thermal expansion correction Rho (T) converted to density field changes; Density field changes are used to calculate the surface subsidence history by the equilibrium settlement equation .

4. The lithospheric thermal structure inversion method based on physical constraints and event recognition according to claim 2, characterized in that, Step S3 includes: Machine learning regression model is used to analyze observation data: ； wherein, is an input feature vector, including the interpolation sequence of temperature evolution history and tectonic subsidence history; is an output parameter vector, including the crustal extension coefficient , the mantle extension coefficient , the event start time , the duration , and the event validity identifier ; Training data set is generated, including: Event parameter random sampling: in the preset geological range, a plurality of period tensile event parameters are randomly generated, including event period, start time of each event, duration, crustal tensile coefficient and mantle tensile coefficient ;​​​ Forward modeling generates response: input multi-stage extension event parameters to step S2, calculate the corresponding temperature evolution history and tectonic subsidence history ; Add observation noise: In the generated temperature history and the constructed subsidence history superimposed with normally distributed random noise; Constructing training samples: the noise-added temperature evolution history and tectonic subsidence history sequence are taken as input features , the original event parameter group for generating the data is taken as output label , which together constitute a training sample ; training a multi-output regression model to obtain a mapping function , to establish a mapping relationship from the observation data to the inversion parameters; The target function is generated as: ； wherein, is a mathematical expectation operator; is a real event parameter vector known in the synthetic data; is a predicted output of the machine learning model for the synthetic input features ; is a norm squared; The intelligent estimated values of the number of events, extension coefficient and time parameters are output.

5. The lithospheric thermal structure inversion method based on physical constraints and event recognition according to claim 4, characterized in that, Step S4 includes: Based on the intelligent estimated values of step S3, differential evolution algorithm is used for global optimization in the parameter search space of step S1, including: Temperature evolution history and tectonic subsidence history input in step S1 are taken as fitting targets, and multi-objective loss function is constructed: ； Wherein: Temperature fitting error: ； Subsidence fitting error: ； Coupling penalty term: ； in, The value of the multi-objective loss function; These are the weighting coefficients; This is the penalty coefficient; This represents the temperature fitting error in the temperature history. This represents the number of temperature observation data points. For the first One observed temperature value; For the first One simulated temperature value; The settlement fitting error in the settlement history; This represents the number of settlement observation data points; For the first One observed settlement value; For the first One simulated settlement value; To calculate the crustal extension coefficient for all identified tectonic events. With mantle stretching coefficient The sum of squares of the differences.

6. The lithospheric thermal structure inversion method based on physical constraints and event recognition according to claim 4, characterized in that, Step S5 includes: Physical consistency test of inversion result is carried out: ； wherein, is a physical consistency integrated score, the higher the score, the more consistent the inversion result is with the preset physical constraints; is a weight of the th physical constraint; is a satisfaction degree score of the th physical constraint; Monte Carlo method is used to evaluate parameter uncertainty: ； wherein, is the posterior standard deviation of the inverted parameter; is the number of effective samples obtained from the Monte Carlo sampling; is the value of the parameter in the i-th Monte Carlo sample; is the average value of the parameter over all samples. A plurality of equivalent solutions are generated, and the degree of multi-solution of the inversion result is analyzed; the final inversion parameters and confidence interval are output.

Images