Method and device for predicting multiple parameters of semi-enclosed in-situ conversion of low to medium mature shale oil

Abstract

The application provides a kind of middle-low mature shale oil semi-enclosed in-situ conversion multi-parameter coupling prediction method and device, which can be used in oil and gas field development technical field, the method comprises the following steps: constructing a molecular model suitable for middle-low mature shale oil components; setting the dynamic pressure regulation boundary of semi-enclosed system, carrying out in-situ conversion simulation based on reaction molecular dynamics; combining Arrhenius formula to calculate pyrolysis kinetic parameters, and constructing a multi-working condition simulation sample set; training a deep learning model to realize rapid prediction of product characteristics and kinetic parameters, and completing parameter optimization and mechanism analysis. The application solves the problems of inconsistency between traditional simulation system and actual formation environment, low efficiency of kinetic parameter acquisition, and difficulty in multi-parameter coupling prediction, realizes accurate characterization and rapid prediction from micro thermal evolution to macro parameter, and provides reliable technical support for in-situ conversion process design and optimization.

Description

Technical Field

[0001] This invention relates to the field of shale oil in-situ conversion and development technology, and in particular to a multi-parameter coupled prediction method and apparatus for semi-closed in-situ conversion of medium- and low-maturity shale oil. Background Technology

[0002] Medium- and low-maturity shale oil, as an important component of unconventional oil and gas resources, is characterized by abundant reserves and wide distribution. Its efficient development is of great significance for alleviating the contradiction between energy supply and demand. In-situ conversion technology for medium- and low-maturity shale oil is the core means to achieve its large-scale development. This technology injects heat into underground shale reservoirs to induce pyrolysis of the organic matter in the shale, generating recoverable resources such as gaseous hydrocarbons and light oils. However, because the in-situ conversion process of shale oil involves complex physicochemical changes and is affected by multiple factors such as temperature, pressure, composition, and heating rate, traditional simulation and experimental methods are insufficient to accurately characterize its conversion mechanism and laws. This results in long optimization cycles and high costs for process parameters, severely restricting the engineering application of in-situ conversion technology.

[0003] In existing technologies, in-situ shale oil conversion simulations mainly employ open or closed systems. In open system simulations, the oil and gas molecules generated by pyrolysis continuously escape, causing the system pressure to become unstable, which is significantly inconsistent with the actual dynamic pressure equilibrium conditions in formations. In closed system simulations, oil and gas molecules cannot escape, and the system pressure continuously increases, similarly failing to replicate the real development environment. The simulation results of both systems deviate significantly from actual engineering conditions. Furthermore, pyrolysis kinetic parameters (such as activation energy and pre-exponential factor) are core indicators reflecting the conversion characteristics of shale oil. Existing methods primarily rely on indoor thermal simulation experiments, which require significant time for sample preparation, experimental operations, and data analysis. Moreover, the experimental conditions involve the loss of light hydrocarbons, limiting the accuracy of the results. While reaction molecular dynamics simulation technology can capture molecular evolution processes at the microscopic level, single-condition simulations can take days or even weeks, making it difficult to meet the needs of multi-parameter coupled optimization. Furthermore, the existing technology lacks a unified quantitative calculation standard for shale oil cracking rate, resulting in a lack of comparability between different research results; it also fails to fully consider the comprehensive impact of various factors on the conversion effect, making it impossible to achieve rapid and accurate prediction of product characteristics and kinetic parameters, leading to low engineering conversion efficiency of simulation results.

[0004] Therefore, developing a method and system that can accurately simulate the dynamic pressure environment of real formations, clarify the calculation methods of core indicators, comprehensively consider key process parameters, quickly obtain pyrolysis kinetic parameters, and realize multi-parameter coupled prediction has become an urgent need for the development of in-situ conversion technology for medium and low maturity shale oil.

[0005] This section is intended to provide background or context for the embodiments of this application set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0006] One objective of this invention is to provide a multi-parameter coupled prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil. By constructing a semi-closed dynamic pressure-controlled molecular simulation system, it achieves accurate replication of the actual formation environment, significantly improving the simulation realism and reliability of the in-situ conversion process. Through coupling reaction kinetic equations and deep learning models, it realizes rapid calculation and multi-parameter coupled prediction of pyrolysis kinetic parameters, greatly improving parameter acquisition efficiency and prediction accuracy. It can comprehensively consider the coupling effects of key process parameters, providing efficient and scientific technical support for the optimized design of shale oil in-situ conversion processes. Another objective of this invention is to provide a multi-parameter coupled prediction device for semi-closed in-situ conversion of medium- and low-maturity shale oil. A further objective of this invention is to provide a computer-readable medium. A final objective of this invention is to provide a computer device.

[0007] To achieve the above objectives, this invention discloses a multi-parameter coupled prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil, comprising: Multiple medium- and low-maturity shale oil samples were collected based on thermal pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil was constructed based on kerogen molecules and selected light oil molecules. Based on the pre-set boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, the molecular dynamics simulation of the deheating reaction of the target shale oil three-dimensional periodic molecular model is carried out to obtain a multi-condition parameter set, which includes time-series parameters and product composition parameters. The simulated yield and simulated shale oil cracking rate of each product are calculated based on the product composition parameters. The simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius formula and constructing a multi-condition simulation sample set. By using a multi-condition simulation sample set, a model is built on a pre-defined fully connected neural network to generate a multi-parameter coupled prediction model. By using a multi-parameter coupled prediction model, parameter prediction is performed based on the operating conditions to be predicted, generating the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

[0008] Preferably, multiple medium- and low-maturity shale oil samples were collected based on pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil was constructed based on kerogen molecules and selected light oil molecules, including: Thermal pyrolysis gas chromatography-mass spectrometry analysis was performed on various medium- and low-maturity shale oil samples to obtain hydrocarbon product composition data for different medium- and low-maturity shale oil samples; Based on the component data of kerogen molecules and selected light oil molecules in the hydrocarbon product composition data, a simulated mass ratio range is generated. Kerogen molecules and light oil molecules were mixed separately according to different simulated mass ratios to construct initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios. The initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios were structurally optimized and annealed to generate the target three-dimensional periodic molecular model of shale oil.

[0009] Preferably, the simulation parameters include: heating rate, initial temperature, and final reaction temperature; The boundary conditions of a semi-closed system include: boundary response parameters and boundary response rules; Boundary parameters include: pressure threshold; The boundary response rule is as follows: when the system pressure is greater than or equal to the preset pressure threshold, gaseous hydrocarbon molecules with a carbon atom number within the preset range are allowed to escape through the boundary; when the system pressure is less than the pressure threshold, hydrocarbon molecules are not allowed to escape.

[0010] Preferably, based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, a molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model is performed to obtain a multi-condition parameter set, including: Based on different simulation parameters, the molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model was carried out multiple times using the reaction force field. The escape of gaseous hydrocarbon molecules was controlled according to the boundary response rule. When the system pressure is greater than or equal to the pressure threshold, the system pressure is controlled to maintain dynamic equilibrium. Monitor and record the timing parameters and product composition parameters during the simulation process to generate a multi-condition parameter set.

[0011] Preferably, the simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius formula, and a multi-condition simulation sample set is constructed, including: The first-order reaction kinetic equation was coupled with the Arrhenius equation, and the simulated pyrolysis kinetic parameters were calculated based on the simulated shale oil cracking rate and time series parameters. Based on the simulated yield of each product, the simulated shale oil cracking rate, and the simulated pyrolysis kinetic parameters under different operating conditions, a multi-condition simulation sample set was constructed.

[0012] Preferably, a multi-parameter coupled prediction model is generated by constructing a model on a pre-defined fully connected neural network using a multi-condition simulation sample set, including: The multi-condition simulation sample set is divided into a training sample set, a validation sample set, and a test sample set. Using a pre-defined adaptive optimizer and loss function, the fully connected neural network is iteratively trained and its model parameters are optimized based on the training sample set. The training process is monitored using a validation sample set based on an early stopping strategy to build an initial prediction model. The performance of the initial prediction model was tested using a test sample set. If the test passes, the initial prediction model will be determined as a multi-parameter coupled prediction model; If the test fails, repeat the steps of dividing the multi-condition simulation sample set into a training sample set, a validation sample set, and a test sample set.

[0013] Preferably, a multi-parameter coupled prediction model is used to predict parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product, including: The operating parameters to be predicted are normalized to obtain the input features to be predicted. The input features to be predicted are input into a multi-parameter coupled prediction model and forward propagated to generate normalized prediction output features. The normalized predicted output features are denormalized to generate the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

[0014] This invention also discloses a semi-closed in-situ conversion multi-parameter coupled prediction device for medium- and low-maturity shale oil, comprising: The molecular model construction module is used to construct a three-dimensional periodic molecular model of the target shale oil based on various medium- and low-maturity shale oil samples collected by thermal pyrolysis gas chromatography-mass spectrometry analysis, and based on kerogen molecules and selected light oil molecules. The dynamic simulation execution module is used to perform molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model based on the preset boundary conditions, simulation parameters and selected reaction force field of the semi-closed system, and obtain a multi-condition parameter set, which includes time parameters and product composition parameters. The sample set construction module is used to calculate the simulated yield and simulated shale oil cracking rate of each product based on the product composition parameters, and to couple the first-order reaction kinetic equation with the Arrhenius formula to obtain the simulated pyrolysis kinetic parameters and construct a multi-condition simulation sample set. The deep learning training module is used to build a model on a pre-defined fully connected neural network using a multi-condition simulated sample set, and generate a multi-parameter coupled prediction model. The parameter prediction module is used to predict parameters based on the operating conditions to be predicted through a multi-parameter coupled prediction model, and generate the target yield, target shale oil cracking rate and target pyrolysis kinetic parameters for each product.

[0015] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0016] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0017] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0018] This invention utilizes various medium- and low-maturity shale oil samples acquired through pyrolysis gas chromatography-mass spectrometry (GC-MS) analysis. A three-dimensional periodic molecular model of the target shale oil is constructed based on kerogen molecules and selected light oil molecules. Based on pre-defined boundary conditions, simulation parameters, and selected reaction force fields of the semi-closed system, molecular dynamics simulations of the depyrolysis reaction in the three-dimensional periodic molecular model of the target shale oil are performed to obtain a multi-condition parameter set, including time-series parameters and product composition parameters. The simulated yield of each product and the simulated shale oil cracking rate are calculated based on the product composition parameters. The simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius equation, thus constructing a multi-condition simulation sample set. Using this multi-condition simulation sample set, a pre-defined fully connected... A neural network is used to build a model, generating a multi-parameter coupled prediction model. This model predicts parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product. By constructing a semi-closed, dynamically pressure-controlled molecular simulation system, accurate replication of the actual formation environment is achieved, significantly improving the simulation realism and reliability of the in-situ conversion process. Furthermore, by coupling reaction kinetic equations with a deep learning model, rapid calculation and multi-parameter coupled prediction of pyrolysis kinetic parameters are realized, greatly improving parameter acquisition efficiency and prediction accuracy. This approach comprehensively considers the coupling effects of key process parameters, providing efficient and scientific technical support for the optimized design of shale oil in-situ conversion processes. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a multi-parameter coupled prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil provided in an embodiment of the present invention; Figure 2 A flowchart of another method for multi-parameter coupled prediction of semi-closed in-situ conversion of medium- and low-maturity shale oil provided in this embodiment of the invention; Figure 3 A molecular structure diagram of kerogen and light oil molecules provided in an embodiment of the present invention; Figure 4 A schematic diagram illustrating the simulation results of pressure regulation in a semi-closed system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the variation of the yield of each component with temperature and time, provided as an embodiment of the present invention. Figure 6 A schematic diagram of a fully connected neural network architecture provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the loss value during actual training, provided as an embodiment of the present invention. Figure 8 This is a schematic diagram of a semi-closed in-situ conversion multi-parameter coupled prediction device for medium- and low-maturity shale oil provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To facilitate understanding of the technical solutions provided in this application, the relevant content of the technical solutions will be described below. This invention is applicable to the simulation, pyrolysis kinetic parameter calculation, product characteristic prediction, and process parameter optimization of in-situ conversion processes of medium- and low-maturity shale oil containing kerogen and light oil molecules under dynamic pressure control. It achieves accurate simulation of the in-situ conversion process of medium- and low-maturity shale oil, rapid calculation of pyrolysis kinetic parameters, and efficient prediction and design of product characteristics. It solves technical problems such as the inconsistency between traditional simulation systems and actual semi-closed geological conditions, and the difficulty in accurately predicting key process parameters, providing scientific and reliable technical support for the engineering application of in-situ conversion technology for medium- and low-maturity shale oil.

[0023] The following example uses a semi-closed in-situ conversion multi-parameter coupling prediction device for medium- and low-maturity shale oil as the execution subject to illustrate the implementation process of the multi-parameter coupling prediction method for medium- and low-maturity shale oil provided in this embodiment of the invention. It is understood that the execution subject of the multi-parameter coupling prediction method for medium- and low-maturity shale oil provided in this embodiment of the invention includes, but is not limited to, a semi-closed in-situ conversion multi-parameter coupling prediction device for medium- and low-maturity shale oil.

[0024] Figure 1 A flowchart of a multi-parameter coupled prediction method for semi-closed in-situ transformation of medium- and low-maturity shale oil, provided by an embodiment of the present invention, is shown below. Figure 1 As shown, the method includes: Step 101: Analyze various medium- and low-maturity shale oil samples collected by pyrolysis gas chromatography-mass spectrometry (Py-GC / MS), and construct a three-dimensional periodic molecular model of the target shale oil based on kerogen molecules and selected light oil molecules.

[0025] In this embodiment of the invention, Py-GC / MS analysis was performed on medium- and low-maturity shale oil samples to obtain hydrocarbon product composition data for different samples. Specifically, the relative contents of hydrocarbons with different carbon numbers were identified and quantified from the experimental spectra, thereby screening out light oil molecules with a high proportion in the products as representative components, ensuring that the constructed initial three-dimensional periodic molecular model of shale oil can truly reflect the chemical characteristics of actual shale oil.

[0026] Kerogen molecules with specific structures were selected and mixed with the aforementioned light oil molecules according to a preset mass ratio to construct an initial three-dimensional periodic molecular model. This process was completed using molecular simulation software, which filled the periodic boxes with molecules of different components, allowing the model to replicate the organic matter composition of actual shale at a microscopic scale.

[0027] Furthermore, to eliminate any unreasonable conformations or local high-energy sites that may arise during the construction of the initial three-dimensional periodic molecular model of shale oil, structural optimization and simulated annealing are required. Structural optimization adjusts atomic coordinates using an energy minimization algorithm to achieve a locally stable state. Simulated annealing, on the other hand, controls heating-cooling cycles to fully relax the molecular chains and allow them to cross energy barriers, ultimately obtaining a thermodynamically stable target three-dimensional periodic molecular model of shale oil.

[0028] This step, through meticulous component screening and model building, ensures that the initial state of the subsequent pyrolysis simulation is highly consistent with the actual sample, significantly improving the authenticity and reliability of the simulation results.

[0029] Step 102: Based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, perform molecular dynamics simulation of the deheating reaction of the target shale oil three-dimensional periodic molecular model to obtain a multi-condition parameter set.

[0030] In this embodiment of the invention, boundary conditions and simulation parameters of a semi-closed system are set based on a three-dimensional periodic molecular model of the target shale oil. The core of this operation is defining the physical constraints of the simulation environment: including setting thermodynamic parameters such as the final reaction temperature and heating rate, and constructing boundary response rules based on actual geological conditions. Specifically, a pressure threshold needs to be preset as a benchmark for system pressure control, and the selective escape conditions of gaseous hydrocarbon molecules need to be clearly defined, for example, only allowing small hydrocarbon molecules with 4 or fewer carbon atoms to escape through the boundary.

[0031] A force field suitable for the pyrolysis reaction of shale oil was selected, and molecular dynamics simulations of the semi-closed system pyrolysis reaction were conducted on a three-dimensional periodic molecular model of the target shale oil. During the simulation, the program monitored the pressure changes of the system in real time: when the real-time system pressure reached or exceeded the set pressure threshold, the gaseous hydrocarbon molecule escape mechanism was automatically activated, allowing small hydrocarbon molecules of a certain size to escape through the boundary, thereby causing the system pressure to drop and be maintained within the dynamic equilibrium range near the threshold; when the pressure was below the threshold, the escape mechanism was closed, and all molecules were confined within the system to participate in further reactions. This dynamic regulation mechanism accurately replicated the semi-closed characteristics of real formations where fluid discharge and formation reached equilibrium.

[0032] In this embodiment of the invention, rich simulation samples are generated by adjusting the boundary conditions and simulation parameters of the semi-closed system. Specifically, the four key parameters—the mass ratio of kerogen to light oil molecules, the pressure threshold of the semi-closed system, the final reaction temperature, and the heating rate—are changed to form multiple different combinations of operating conditions. Steps 101 and 102 are repeated for each combination, and the simulation process records the molecular structure evolution trajectory and product generation data in real time throughout the simulation, thereby obtaining raw simulation data covering a wide parameter range. After the simulation, a multi-condition parameter set is extracted from the raw simulation data. This multi-condition parameter set includes time-series parameters and product composition parameters. Time-series parameters include, but are not limited to, macroscopic thermodynamic information such as system temperature, pressure, and total number of molecules at each moment; product composition parameters include, but are not limited to, the molecular formulas and relative mass distributions of gaseous hydrocarbons, light oils, heavy oils, and coke pitch corresponding to different carbon atom number ranges.

[0033] This step addresses the problem that traditional closed or open systems cannot accurately reflect the formation pressure environment by introducing semi-closed boundary conditions with dynamic pressure regulation, significantly improving the engineering applicability of the simulation results. Simultaneously, the real-time recorded multi-dimensional parameters provide high-quality foundational data for subsequent dynamic calculations and dataset construction.

[0034] Step 103: Calculate the simulated yield and simulated shale oil cracking rate of each product based on the product composition parameters, and obtain the simulated pyrolysis kinetic parameters by fitting the first-order reaction kinetic equation with the Arrhenius formula, and construct a multi-condition simulation sample set.

[0035] In this embodiment of the invention, for each set of operating conditions and its product composition parameters, the simulated yields of various products are calculated according to a preset formula, including gaseous hydrocarbon yield, light oil yield, heavy oil yield, and coke pitch yield. The simulated shale oil cracking rate is also calculated as a core indicator for evaluating conversion efficiency. Based on this, assuming the pyrolysis reaction conforms to first-order reaction kinetics, the first-order reaction kinetic equation is coupled with the Arrhenius equation to derive a linear fitting equation. Using characteristic temperature points extracted from the time-series parameters and their corresponding conversion rate data, the simulated pyrolysis kinetic parameters, namely the activation energy and pre-exponential factor, are obtained through linear regression fitting.

[0036] After completing the above calculations, a multi-condition simulation sample set is constructed: the mass ratio of kerogen to light oil, pressure threshold, final reaction temperature, and heating rate of each condition are defined as 4-dimensional input features; the gaseous hydrocarbon yield, light oil yield, heavy oil yield, coke pitch yield, shale oil cracking rate, activation energy, and pre-exponential factor are defined as 7-dimensional output labels, forming a complete "input-output" sample pair.

[0037] Furthermore, the multi-condition simulation sample set is preprocessed: outlier samples such as those with unsatisfactory linear fitting coefficients are removed to ensure that the number of effective samples meets the preset requirements; then the input features and output labels are normalized, and the Min-Max method is used to map all data to the [0,1] interval to eliminate the influence of different physical dimensions on subsequent model training.

[0038] This step constructs a high-quality multi-condition simulation sample set through multi-condition combination and standardized data processing, providing a reliable data foundation for deep learning models; at the same time, it realizes the rapid calculation of pyrolysis kinetic parameters by coupling kinetic equations, which greatly improves the efficiency of parameter acquisition.

[0039] Step 104: Using a multi-condition simulation sample set, construct a model for the preset fully connected neural network to generate a multi-parameter coupled prediction model.

[0040] In this embodiment of the invention, the multi-condition simulation sample set is divided into a training set, a validation set, and a test set according to a preset ratio. Specifically, the samples are randomly split in a ratio of 7:1.5:1.5: the training set is used for iterative updates of model parameters, the validation set is used to monitor the training process and adjust hyperparameters, and the test set is stored independently for the final evaluation of the model's generalization ability. This division ensures the objectivity and reliability of model training, tuning, and performance evaluation.

[0041] A fully connected neural network was constructed, employing a three-tiered structure of "input layer - hidden layer - output layer". The input layer has 4 neurons, strictly corresponding to the input feature dimensions (kerogen to light oil mass ratio, pressure threshold, final reaction temperature, and heating rate). The hidden layer consists of two layers: the first hidden layer has 90 neurons, and the second has 45, both using the ReLU activation function to introduce nonlinear transformations, enabling the network to learn the complex coupling relationship between input features and output labels. The output layer has 7 neurons, consistent with the output label dimensions (yield of each product, shale oil cracking rate, activation energy, and pre-exponential factor), and uses the Linear activation function to directly output continuous numerical prediction results.

[0042] The formula for the ReLU activation function is: ReLU(x)=max(0,x). This function can effectively solve the gradient vanishing problem in deep networks, while introducing nonlinearity, enabling the model to fit complex parameter coupling relationships.

[0043] Before training the model, key training parameters need to be configured. One possible approach is to use mean squared error as the loss function to measure the deviation between the predicted and actual values; select the Adam adaptive optimizer with an initial learning rate of 0.001, a momentum decay coefficient of 0.9, and an adaptive learning rate decay coefficient of 0.999 to accelerate convergence and stabilize the training process; set an upper limit of 800 iterations and introduce an early stopping strategy—monitor the validation set loss after each training round, and if the loss no longer decreases after 8 consecutive rounds, terminate training early to effectively avoid model overfitting.

[0044] During training, training data is input into the network in batches, forward propagation is performed to calculate predicted values, the error is calculated using a loss function, and then the network weights and biases are updated based on the backpropagation algorithm. After a certain number of rounds, the performance of the current model is evaluated using a validation set, and hyperparameters are dynamically adjusted or early stopping is triggered based on the validation set loss. After training is complete, the model with the lowest validation set loss is saved as the final multi-parameter coupled prediction model.

[0045] Finally, the trained model was evaluated using an independent test set, and the mean absolute error and coefficient of determination R were calculated. 2 And the maximum relative error. The evaluation results are required to meet the preset thresholds: mean absolute error ≤ 0.005, R0 2 ≥0.97, maximum relative error ≤3%. Validation on the test set ensures the model has good prediction accuracy and generalization ability, and can be reliably applied to rapid prediction of unknown working conditions.

[0046] This step, through a carefully designed neural network architecture and training strategy, achieves in-depth mining of multi-parameter coupling relationships, significantly improving the prediction accuracy of product characteristics and dynamic parameters. At the same time, the early stopping strategy and validation set monitoring mechanism effectively prevent overfitting, ensuring the model's generalization performance and laying a solid model foundation for subsequent rapid prediction and process optimization.

[0047] Step 105: Using a multi-parameter coupled prediction model, perform parameter prediction based on the operating parameters to be predicted, and generate the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

[0048] In this embodiment of the invention, the operating parameters to be predicted are obtained, namely: a set of known 4-dimensional input features, including the mass ratio of kerogen to light oil molecules, pressure threshold, final reaction temperature, and heating rate. Since the deep learning model is trained on normalized data, the input parameters to be predicted need to undergo Min-Max normalization processing, which is completely consistent with step 103, to map them to the [0, 1] interval, eliminate dimensional differences, and ensure that the input data conforms to the model's acceptable format.

[0049] Load the trained multi-parameter coupled prediction model file, which contains the network weights, biases, and model architecture information obtained in step 104. Substitute the normalized input features into the model and calculate directly using the forward propagation formula: the input data sequentially passes through the input layer, two hidden layers (each layer first undergoes a linear transformation, then introduces non-linearity through the ReLU activation function), and the output layer (linear activation), finally outputting a normalized 7-dimensional prediction value. This calculation process is a single forward computation, requiring no iterative training or repeated simulations, and can therefore be completed in milliseconds.

[0050] The normalized predicted values ​​of the model output are subjected to Min-Max inverse normalization. Using the original minimum and maximum values ​​of each output label saved in step 103, the predicted values ​​are restored to the actual physical quantities, resulting in 7-dimensional output results: gaseous hydrocarbon yield, light oil yield, heavy oil yield, coke pitch yield, shale oil cracking rate, activation energy, and pre-exponential factor. These results are the target yields (gaseous hydrocarbon yield, light oil yield, heavy oil yield, coke pitch yield), target shale oil cracking rate, and target pyrolysis kinetic parameters (activation energy and pre-exponential factor), which can be directly used for engineering analysis and decision-making.

[0051] This step, by encapsulating the trained deep learning model, achieves second-level rapid prediction of unknown operating conditions, fundamentally solving the pain point that traditional molecular dynamics simulations for single operating conditions take several days and cannot meet the needs of rapid engineering iteration. Simultaneously, the complete data preprocessing and post-processing workflow ensures the physical interpretability and engineering applicability of the prediction results, providing efficient and reliable technical support for parameter optimization and real-time decision-making in in-situ conversion processes.

[0052] In the technical solution provided by this invention, multiple medium- and low-maturity shale oil samples are collected based on pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil is constructed based on kerogen molecules and selected light oil molecules. Based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, molecular dynamics simulation of the depyrolysis reaction of the target shale oil is performed on the three-dimensional periodic molecular model, resulting in a multi-condition parameter set, which includes time-series parameters and product composition parameters. The simulated yield and simulated shale oil cracking rate of each product are calculated based on the product composition parameters, and the simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius equation, thus constructing a multi-condition simulation sample set. Through the multi-condition simulation sample set, the... A pre-defined fully connected neural network is used to construct the model, generating a multi-parameter coupled prediction model. This model predicts parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product. By constructing a semi-closed, dynamically pressure-controlled molecular simulation system, accurate replication of the actual formation environment is achieved, significantly improving the simulation realism and reliability of the in-situ conversion process. Furthermore, by coupling reaction kinetic equations with a deep learning model, rapid calculation and multi-parameter coupled prediction of pyrolysis kinetic parameters are realized, greatly improving parameter acquisition efficiency and prediction accuracy. This approach comprehensively considers the coupling effects of key process parameters, providing efficient and scientific technical support for the optimized design of shale oil in-situ conversion processes.

[0053] Figure 2 A flowchart of another method for multi-parameter coupled prediction of semi-closed in-situ conversion of medium- and low-maturity shale oil provided in this embodiment of the invention is shown below. Figure 2 As shown, the method includes: Step 201: Perform Py-GC / MS analysis on various medium- and low-maturity shale oil samples to obtain hydrocarbon product composition data for different medium- and low-maturity shale oil samples.

[0054] In this embodiment of the invention, each step is performed by a semi-closed in-situ conversion multi-parameter coupled prediction device for medium- and low-maturity shale oil.

[0055] Specifically, each shale oil sample was subjected to programmed temperature pyrolysis in an anaerobic environment, causing the organic matter in the sample to decompose into volatile fragment molecules. These fragments were then carried by a carrier gas into a gas chromatography column for separation, with molecules of different components eluting sequentially due to their different retention times in the column. Subsequently, these fragments entered a mass spectrometer, where they were ionized by electron bombardment to generate characteristic fragment ions. Mass spectrometry analysis yielded the corresponding mass spectra for each component. By comparing the characteristics of standard spectral libraries and analytical spectra, the specific hydrocarbon molecules contained in the pyrolysis products could be qualitatively identified, such as alkanes, alkenes, and aromatic hydrocarbons with different carbon numbers. Simultaneously, based on the peak area or peak height of the chromatographic peaks, combined with correction factors, quantitative calculations were performed to obtain the relative content of each hydrocarbon component. Finally, the qualitative and quantitative results were integrated to form the hydrocarbon product composition data for the sample.

[0056] This step, through detailed experimental analysis of real samples, obtained representative information on the molecular components and structure of light oil, providing a reliable data foundation for the subsequent construction of a molecular model that can truly reflect the characteristics of actual shale oil, and fundamentally ensuring the representativeness of the samples and the credibility of the results in the simulation study.

[0057] Step 202: Based on the component data of kerogen molecules and selected light oil molecules in the hydrocarbon product composition data, generate a simulated mass ratio range.

[0058] In this embodiment of the invention, carbon atoms within a specific range (C) are selected from the component composition data. a To C b The light oil molecules within the range of ) are used as the light oil molecules required for subsequent modeling. Among them, the value of 'a' ranges from 5 to 8, and the value of 'b' ranges from 10 to 13.

[0059] As an alternative, carbon atoms with numbers ranging from C5 to C6 are selected. 13 The medium hydrocarbon fragments, as representative light oil molecules, specifically include benzene (C6H6, relative molecular mass 78) and n-heptane (C7H6). 16 (Relative molecular mass 100), nonane (C9H) 20 (relative molecular mass 128) and undecene (C 11 H 22 (Relative molecular mass 154). Light oil molecules in this range are the main organic components in medium- and low-maturity shale oils, excluding kerogen, and have a significant impact on the pyrolysis behavior of in-situ conversion processes.

[0060] Simultaneously, it is necessary to perform kerogen fine preparation on medium- and low-maturity shale oil samples, that is, to remove minerals from the samples through chemical methods such as acid treatment and enrich them to obtain pure kerogen solids. At the same time, Soxhlet extraction is used to separate and extract soluble organic matter from the samples to obtain light oil molecular components. By weighing the separated products, the actual mass ratio of kerogen to light oil molecules in different samples can be calculated.

[0061] Based on the combined measurements of multiple samples, the preset mass ratio range of kerogen molecules to light oil molecules was determined to be k1:1 to k2:1, where k1 ranges from 5 to 8 and k2 ranges from 15 to 20. As an alternative approach, a typical mass ratio range of 5:1 to 20:1 was selected. This range covers the common ratio range of the two components in medium-to-low maturity shale oil and has good sample representativeness.

[0062] When calculating the mass ratio, the following formula is used for conversion: Mass ratio of kerogen to light oil molecules = (Number of kerogen molecules × their relative molecular mass) : (Number of light oil molecules × their relative molecular mass). Figure 3 A molecular structure diagram of kerogen and light oil molecules provided in an embodiment of the present invention is shown below. Figure 3 As shown, dark gray atoms represent carbon (C) atoms, light gray atoms represent hydrogen (H) atoms, red atoms represent oxygen (O) atoms, blue atoms represent nitrogen (N) atoms, and yellow atoms represent sulfur (S) atoms. The kerogen molecule uses the specific molecular formula Ck. 202 H 230 N6O 14 S4, the light oil molecules adopt the C6H6 and C7H6 selected above. 16 C9H 20 And C 11 H 22 The relative molecular masses of representative molecules are weighted averaged or calculated separately.

[0063] This step, through pretreatment and quantitative analysis of different medium- and low-maturity shale oil samples, obtained the actual mass ratio range of kerogen and light oil molecules, providing a scientific basis for the subsequent construction of molecular models with different component ratios. At the same time, the clear mass ratio range ensures that the constructed molecular model can represent the organic matter composition of the actual reservoir, significantly improving the representativeness and engineering applicability of the simulation study.

[0064] Step 203: Mix kerogen molecules and light oil molecules according to different simulated mass ratios to construct initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios.

[0065] In this embodiment of the invention, taking a mass ratio range of 5:1 to 20:1 as an example, representative specific ratio values ​​are selected for modeling. In this embodiment, four typical ratios are selected, namely, the mass ratios of kerogen to light oil molecules are 5:1, 10:1, 15:1 and 20:1, to cover the entire range and examine the influence of different component compositions on pyrolysis behavior.

[0066] The model was built using tools in molecular simulation software for constructing amorphous three-dimensional periodic structures. The specific procedure involved converting kerogen and light oil molecules into a corresponding molecular number ratio according to a preset mass ratio. For example, for a 5:1 mass ratio, the specific number of kerogen molecules and various light oil molecules required was calculated using the following formula: (Number of kerogen molecules × relative molecular mass of kerogen molecules) : (Number of light oil molecules × relative molecular mass of light oil molecules) = 5 : 1.

[0067] In a tool for constructing amorphous three-dimensional periodic structures, the initial size of the three-dimensional periodic box is set, and a calculated number of kerogen and light oil molecules are randomly placed into the box. The tool automatically adjusts the initial positions and orientations of the molecules to avoid severe overlap between atoms, thereby generating a reasonable initial three-dimensional periodic molecular model of shale oil. This model has periodic boundary conditions in the x, y, and z directions, meaning the model is equivalent to an infinitely repeating basic unit, which can more realistically simulate the continuous medium characteristics of shale oil at the microscopic scale.

[0068] For each selected simulation mass ratio (5:1, 10:1, 15:1, and 20:1), the above modeling process is repeated to obtain initial molecular models for different simulation mass ratios, providing diverse initial states for subsequent pyrolysis simulations.

[0069] This step systematically constructs molecular models with different component ratios, comprehensively reflecting the component diversity of medium- and low-maturity shale oil in actual reservoirs, and providing a foundation for subsequent research on the influence of the kerogen to light oil ratio on the in-situ conversion process. At the same time, the introduction of a three-dimensional periodic structure eliminates boundary effects, making the simulation results closer to the molecular environment in actual formations.

[0070] Step 204: Optimize and anneal the initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios to generate the target three-dimensional periodic molecular model of shale oil.

[0071] In this embodiment of the invention, simulated annealing is used, and the Dreiding force field, which is suitable for organic molecular systems, is selected for energy calculation to eliminate unreasonable conformations and high-energy sites that may be generated during the initial model building process, so as to obtain a thermodynamically stable three-dimensional periodic molecular model of the target shale oil.

[0072] Step 204 specifically includes: Step 2041: Under the NVT ensemble (i.e., an ensemble with constant particle number, volume and temperature), gradually increase the model temperature from room temperature to 800 Kelvin (K) and run the simulation for 200 picoseconds (ps) at this temperature.

[0073] This high-temperature process allows the model to acquire sufficient kinetic energy, enabling the molecular chains to move, rotate, and relax fully, overcoming the local energy barriers in the initial conformation, thereby allowing the entire system to reach a preliminary thermodynamic equilibrium state.

[0074] Step 2042: Switch to the NPT ensemble (i.e., an ensemble with constant particle number, pressure, and temperature), apply a pressure of 30 MPa in the x and y directions of the model, and maintain this pressure condition for 500 ps of simulation.

[0075] This step simulates the horizontal pressure exerted by the overlying rock strata on the formation, allowing the model to be freely adjusted in the vertical direction (z-direction), with the box size undergoing reasonable deformation, gradually approximating the actual pressure state of the formation.

[0076] Step 2043: While keeping the NPT ensemble and 30 MPa pressure constant, gradually decrease the model temperature. Specifically, decrease the temperature by 50 K per step, and maintain the simulation for 500 ps at each temperature point until the temperature drops to 300 K (room temperature).

[0077] This slow cooling process allows the molecular chains enough time to adjust their conformation and gradually move towards a stable arrangement with the lowest energy, avoiding freezing stress or metastable conformations caused by rapid cooling.

[0078] Step 2044: After the temperature drops to 300 K, continue to maintain a pressure of 30 MPa to further shrink the model box until the density ρ of the model reaches the range consistent with the actual kerogen density of the formation, and finally obtain a stable three-dimensional periodic molecular model of the target shale oil for different simulated mass ratios (5:1, 10:1, 15:1, 20:1).

[0079] In this embodiment of the invention, the target density is set to 1.45 grams per cubic centimeter (g / cm³). 3 The allowable deviation range is ±0.05 g / cm³. 3 When the model density falls within this range, the energy minimization process is considered complete. At this point, the model is in a thermodynamically stable equilibrium state and its physical properties closely match those of the actual strata, providing a realistic and reliable initial state for subsequent pyrolysis simulations of semi-closed systems.

[0080] This invention effectively eliminates the influence of human randomness in the model building process by using a simulated annealing and pressure coupling method, ensuring the thermodynamic stability and density accuracy of the molecular model, and laying a solid foundation for the reliability and engineering applicability of subsequent pyrolysis simulation results.

[0081] Step 205: According to different simulation parameters, the reaction force field is used to perform multiple semi-closed system deheating reaction molecular dynamics simulations on the three-dimensional periodic molecular model of the target shale oil. The gaseous hydrocarbon molecules are controlled to escape according to the boundary response rules. When the system pressure is greater than or equal to the pressure threshold, the system pressure is controlled to maintain dynamic equilibrium.

[0082] In this embodiment of the invention, the simulation parameters include: heating rate, initial temperature, and final reaction temperature.

[0083] The heating rate δ ranges from 10 to 30 K / ps. As an optional option, the specific heating rate values ​​are 10 K / ps, 15 K / ps, 20 K / ps, 25 K / ps, and 30 K / ps. This range can achieve a good balance between simulation efficiency and the accuracy of capturing reaction details.

[0084] The initial temperature T0 is uniformly set to 300K, and the final reaction temperature T fin Five characteristic points were selected: 2500K, 3000K, 3500K, 4000K, and 4500K, to cover the complete temperature range of in-situ conversion of medium- and low-maturity shale oil, including key reaction stages such as kerogen pyrolysis and small molecule oil cracking.

[0085] In this embodiment of the invention, the boundary conditions of the semi-closed system include: boundary response parameters and boundary response rules; the boundary response parameters include: pressure threshold; the boundary response rules are: when the system pressure is greater than or equal to the preset pressure threshold, gaseous hydrocarbon molecules with a carbon atom number within a preset range are allowed to escape through the boundary; when the system pressure is less than the pressure threshold, hydrocarbon molecules are not allowed to escape.

[0086] The pressure threshold P0 was set to multiple values ​​of 20 MPa, 25 MPa, 30 MPa, 35 MPa, and 40 MPa to examine the influence of different pressure environments on pyrolysis behavior. The boundary response rule is implemented through the following logic: when the real-time pressure P of the system is greater than or equal to the set pressure threshold P0, and the number of carbon atoms n in the molecule... CWhen the hydrocarbon molecules are within the permissible escape range (C1-C4 in this embodiment, i.e., gaseous hydrocarbon molecules with less than 4 carbon atoms), the gaseous hydrocarbon molecule escape mechanism is activated, allowing these specific small hydrocarbon molecules to escape through the boundary. When the real-time system pressure P is less than the pressure threshold P0, or the number of carbon atoms in the molecules is not within the permissible range, the escape of any molecule is prohibited. Through this rule, the system pressure can be maintained within a stable fluctuation range of P0 ± 0.5 MPa, accurately replicating the semi-closed working condition in real formations where fluid generation and discharge reach dynamic equilibrium.

[0087] For the force field selection, the ReaxFF reaction force field, specifically the C / H / O / N / S / B force field, was chosen to suit the shale oil pyrolysis reaction. This force field can accurately describe the breaking and formation of chemical bonds during the organic matter pyrolysis process. The simulation time step Δt was set to 0.25 femtoseconds (fs) to ensure computational accuracy and numerical stability. Molecular simulation software, such as LAMMPS or AMS, was used for the simulation calculations.

[0088] During the simulation, three-dimensional periodic molecular models of target shale oil with different mass ratios (5:1, 10:1, 15:1, 20:1) were imported into the molecular simulation software, and calculations were started according to the simulation parameters and boundary conditions set above. During the simulation, changes in pressure, temperature, and number of molecules within the system were monitored in real time, and detailed microscopic information such as the coordinates, velocity, chemical bond state, molecular composition, and quantity of each molecule was recorded. Simultaneously, macroscopic thermodynamic parameters such as temperature and pressure of the system were output.

[0089] When the real-time system pressure P reaches or exceeds the pressure threshold P0, a gaseous hydrocarbon molecule escape mechanism is automatically triggered according to preset boundary response rules: through a built-in boundary condition control function, only gaseous hydrocarbon molecules with fewer than 4 carbon atoms (C1-C4) are allowed to escape through the preset boundary, while hydrocarbon molecules with C5 and above remain in the system to continue participating in the reaction. This dynamic regulation process continues to ensure that the system pressure remains stable within the range of P0 ± 0.5 MPa.

[0090] Taking the working conditions of 10:1 molecular weight ratio of kerogen to light oil, 3500K final reaction temperature, 20K / ps heating rate, and 30MPa pressure threshold as an example, Figure 4 A schematic diagram illustrating the simulation results of pressure regulation in a semi-closed system provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the horizontal axis represents simulation time (ps), ranging from 0 to 160, with intervals of 20; the left vertical axis represents the real-time system pressure (MPa), ranging from 0 to 35, with intervals of 5. The solid red line represents the real-time system pressure, and the gray dashed line represents the pressure threshold (30MPa); the right vertical axis represents the cumulative number of C1-C4 gaseous hydrocarbons that have escaped, indicated by blue dots. Figure 4It can be seen that when the real-time pressure of the system reaches and exceeds the pressure threshold (30 MPa), the cumulative number of C1-C4 gaseous hydrocarbons that escape increases sharply.

[0091] Taking the working conditions of 10:1 molecular weight ratio of kerogen to light oil, 3500K final reaction temperature, 20K / ps heating rate, and 30MPa pressure threshold as an example, Figure 5 This is a schematic diagram illustrating the variation of the yield of each component with temperature and time, provided as an embodiment of the present invention. Figure 5 As shown, the horizontal axis represents time (ps), ranging from 0 to 160, with intervals of 10; the vertical axis represents the yield of each component (100%), ranging from 0.0 to 1.0, with intervals of 0.2. The gray solid line represents the yield of C1-C4 gaseous hydrocarbons, and the red solid line represents the yield of C5-C6 gaseous hydrocarbons. 15 Gaseous hydrocarbon yield, the blue solid line represents C 16 -C 40 Gaseous hydrocarbon yield; solid green line represents C 40+ Gaseous hydrocarbon yield. (From...) Figure 5 It can be seen that C 40+ The initial yield of gaseous hydrocarbons was close to 100%, gradually decreasing over time, indicating that macromolecules were continuously broken down and consumed; C 16 -C 40 The gaseous hydrocarbon yield rose briefly in the intermediate phase before rapidly dropping to zero, indicating that it is an unstable intermediate cracking product; C5-C 15 The yield of gaseous hydrocarbons slowly increases from about 60 ps and then tends to stabilize, representing the continuous generation and accumulation of some medium-chain hydrocarbons. The yield of C1-C4 gaseous hydrocarbons is almost zero in the early stage, but rises rapidly after 80 ps, ​​becoming the main product in the later stage, reflecting the large-scale formation of small molecule pyrolysis end products.

[0092] This step addresses the problem that traditional closed or open systems cannot accurately reflect the formation pressure environment by introducing semi-closed boundary conditions with dynamic pressure regulation, significantly improving the agreement between simulation results and actual engineering conditions. Simultaneously, the combination of multiple sets of simulation parameters provides rich foundational data for subsequently constructing a sample set covering a wide range of operating conditions.

[0093] Step 206: Monitor and record the timing parameters and product composition parameters during the simulation process to generate a multi-condition parameter set.

[0094] In this embodiment of the invention, raw data is recorded synchronously during the simulation. When molecular simulation software is used to simulate the molecular dynamics of a pyrolysis reaction in a semi-closed system, molecular structure evolution trajectory files, thermodynamic parameter time series files, and product generation data are generated. The structure evolution trajectory files record the coordinates, velocities, and chemical bond states of all atoms at each time step; the thermodynamic parameter time series files (in this embodiment, in .rkf format) contain macroscopic thermodynamic information such as system temperature, pressure, and energy that change over time; and the product generation data contains the number and type of various molecules present in the system at each time step. These raw files provide complete basic data for subsequent parameter extraction and analysis.

[0095] After the simulation is completed, the parameter extraction stage begins. For molecular models with different simulated mass ratios (5:1, 10:1, 15:1, 20:1) and various combinations of simulation parameters (different pressure thresholds, heating rates, and final reaction temperatures), parameter extraction needs to be performed independently for each operating condition. The extraction process obtains time-series parameters and product composition parameters from recorded time-series and trajectory files.

[0096] The time series parameters include the system temperature T (in K), the real-time system pressure P (in MPa), and the total number of molecules N in the system at each simulation time. These time series parameters reflect the dynamic evolution of the system's thermodynamic state during the pyrolysis reaction.

[0097] The product composition parameters were extracted by analyzing the trajectory file and statistically determining the parameters at the final reaction temperature T. fin At any given time, the number of each type of molecule present in the system is determined, and the molecules are classified according to the number of carbon atoms. Specifically, molecules with C1-C4 carbon atoms are classified as gaseous hydrocarbons, and those with C5-C6 carbon atoms are classified as gaseous hydrocarbons. 15 The molecules are classified as light oils, C 16 -C 40 The molecules are classified as heavy oil, with more than 40 carbon atoms (C). 40+ The molecules or solid residues of these products are classified as coke or pitch. For each type of product, record the specific molecular formulas it contains.

[0098] After completing molecular classification and counting, relative mass calculations are required. All mass-related calculations are based on "relative molecular mass" and are performed by summing the results of "number of molecules × relative molecular mass of the corresponding molecule". For example, the relative mass of a gaseous hydrocarbon is equal to the sum of the number of all C1-C4 molecules multiplied by their respective relative molecular masses. Similarly, the relative masses of light oil, heavy oil, and coke pitch, as well as the total relative mass M of the initial molecular model, can be calculated.

[0099] Through the above processing, a complete parameter set can be obtained for each simulation condition, which includes the time-series parameters (evolution of temperature and pressure over time) and product composition parameters (molecular list and relative mass of various products). By repeating this extraction process for all conditions, a multi-condition parameter set covering different mass ratios, pressure thresholds, final reaction temperatures, and heating rates can be generated, providing standardized basic data for subsequent yield calculations, kinetic parameter fitting, and sample set construction.

[0100] This step, through systematic data recording and parameter extraction, transforms the massive amounts of raw data generated by microscopic molecular simulation into engineering parameters with clear physical meaning, achieving a precise mapping from raw trajectories to quantifiable indicators, and laying a high-quality data foundation for subsequent quantitative analysis and model training.

[0101] Step 207: Calculate the simulated yield and simulated shale oil cracking rate of each product based on the product composition parameters.

[0102] In this embodiment of the invention, the product composition parameters include the relative masses of gaseous hydrocarbons, light oil, heavy oil, and coke pitch. The formulas for calculating the simulated yield of each product and the simulated shale oil cracking rate are as follows: Simulated yield P of gaseous hydrocarbons C1-C4 :

[0103] in, M represents the relative mass of the gaseous hydrocarbon; M is the total relative mass of the initial molecular model.

[0104] Simulated yield P of light oil C5-C15 :

[0105] in, denoted as , where is the relative mass of the light oil; M is the total relative mass of the initial molecular model.

[0106] Simulated yield P of heavy oil C16-C40 :

[0107] in, denoted as ρ, where ρ is the relative mass of the heavy oil; M is the total relative mass of the initial molecular model.

[0108] Simulated yield P of coke and bitumen C40+ :

[0109] in, denoted as ρ, where ρ is the relative mass of coke pitch; M is the total relative mass of the initial molecular model. The simulated yield of coke pitch.

[0110] Shale oil cracking rate Y k Defined as the ratio of kerogen and light oil molecules converted into oil and gas products, simulating the shale oil cracking rate Y. k :

[0111] in, To simulate shale oil cracking rate; .

[0112] Shale oil cracking rate is the ratio of the total relative mass of the three products—gaseous hydrocarbons, light oil, and heavy oil—to the initial total relative mass. Shale oil cracking rate directly reflects the efficiency of converting organic matter in shale oil into recoverable resources.

[0113] Furthermore, in order to fit the pyrolysis kinetic parameters, a temperature-conversion rate dataset needs to be constructed. Specifically, from the time-series parameters extracted in step 206, several representative characteristic temperature points are selected, and the shale oil cracking rate (expressed as a decimal) corresponding to each characteristic temperature point is recorded.

[0114] In this embodiment of the invention, taking the operating conditions of a kerogen to light oil mass ratio of 10:1, a pressure threshold of 30 MPa, a heating rate of 20 K / ps, and a final reaction temperature of 3500 K as an example, 12 characteristic temperature points were extracted from the time-series parameters, namely 2400 K, 2500 K, 2600 K, 2700 K, 2800 K, 2900 K, 3000 K, 3100 K, 3200 K, 3300 K, 3400 K, and 3500 K. For each characteristic temperature point, the shale oil cracking rate Y at that moment was calculated using the product composition data recorded in step 206. k This creates a set of data pairs corresponding to temperature and conversion rate. Example data is shown in Table 1: Table 1

[0115] For example, the shale oil cracking rate is 0.036 at 2400 K, 0.051 at 2500 K, and reaches 0.432 at 3500 K. This dataset fully records the dynamic change of conversion rate with increasing temperature during pyrolysis.

[0116] This step transforms the molecular number data obtained from microscopic simulation into an engineering-significant yield index through a clearly defined quantitative calculation formula, enabling the simulation results to be directly compared with laboratory experiments or field production data. At the same time, the construction of the temperature-conversion dataset provides standardized input data for subsequent coupled reaction kinetic equations, solving activation energy and pre-exponential factors, laying the foundation for kinetic parameter calculation.

[0117] Step 208: Couple the first-order reaction kinetic equation with the Arrhenius equation, and calculate the simulated pyrolysis kinetic parameters based on the simulated shale oil cracking rate and time series parameters.

[0118] In this embodiment of the invention, the time-series parameters include, but are not limited to, the evolution of temperature and pressure over time.

[0119] In this embodiment of the invention, it is assumed that the pyrolysis reaction of shale oil follows first-order reaction kinetics, i.e., the reaction rate is directly proportional to the amount of remaining reactive material. The first-order reaction kinetic equation is:

[0120] in, This simulates the shale oil cracking rate, representing the proportion of converted organic matter; k is the reaction rate constant, measured in seconds. -1 t represents the reaction time required to reach the corresponding characteristic temperature, expressed in seconds. This represents the relative mass of gaseous hydrocarbons. This refers to the relative mass of light oils; This represents the relative mass of heavy oil. This represents the total relative mass of the initial molecular model.

[0121] The pyrolysis kinetic parameters include the shale oil pyrolysis activation energy and the shale oil pyrolysis pre-exponential factor. The relationship between the reaction rate constant k and temperature T is described by the Arrhenius equation, which is:

[0122] Where A is the pre-exponential factor for shale oil pyrolysis, in units of s. -1 E represents the activation energy of shale oil pyrolysis, in J / mol; R is the gas constant, with a value of 8.314 J / (mol). K); T is temperature, in K.

[0123] By coupling the first-order reaction kinetic equation with the Arrhenius formula and integrating over time, the linear fitting equation is derived:

[0124] Where A is the pre-exponential factor for shale oil pyrolysis, in units of s.-1 E represents the activation energy of shale oil pyrolysis, in J / mol; R is the gas constant, with a value of 8.314 J / (mol). K); T is the system temperature in K; t is the reaction time required to reach the corresponding characteristic temperature in seconds; It simulates the cracking rate of shale oil.

[0125] by Using the vertical axis as the ordinate, with Using the x-axis as the horizontal axis, a linear regression was performed using the least squares method to obtain the fitted straight line equation. The activation energy E and the pre-exponential factor A of shale oil pyrolysis were then calculated.

[0126] in, ; ; The slope of the fitted equation; The intercept of the fitted equation is denoted as A; A is the pre-exponential factor of shale oil pyrolysis, in seconds. -1 E represents the activation energy of shale oil pyrolysis, in J / mol; R is the gas constant, with a value of 8.314 J / (mol). K); T is the system temperature in K; t is the reaction time required to reach the corresponding characteristic temperature in seconds; It simulates the cracking rate of shale oil.

[0127] As an optional approach, the predetermined threshold for the coefficient of determination of linear fitting is 0.95 ≤ R. 2 ≤1.0.

[0128] In this embodiment of the invention, taking the working conditions of 10:1 mass ratio of kerogen to light oil, 30 MPa pressure threshold, 20 K / ps heating rate, and 3500 K final reaction temperature as an example, linear fitting was performed based on the example data in Table 1 to obtain the fitting slope. =-2.1×10 4 K, intercept =31.2, substituting into the calculation, we get:

[0129] The coefficient of determination R for linear fitting 2 =0.98, satisfying R 2 The requirement of ≥0.95 indicates that the fitted line matches the data points well, and the obtained activation energy and pre-exponential factor have high reliability.

[0130] This step combines the temperature-conversion data obtained from microscopic simulation with classical reaction kinetics theory, enabling rapid and accurate calculation of pyrolysis kinetic parameters. This avoids problems such as long sample preparation cycles and large losses of light hydrocarbons in traditional experimental methods. At the same time, the linear fitting method and the coefficient of determination test ensure the objectivity and reliability of the parameter solution, providing key kinetic basis for subsequent process optimization.

[0131] Step 209: Construct a multi-condition simulation sample set based on the simulated yield of each product, the simulated shale oil cracking rate, and the simulated pyrolysis kinetic parameters under different working conditions.

[0132] In this embodiment of the invention, a multi-condition combination scheme is designed to systematically cover the key influencing factors of in-situ conversion of medium- and low-maturity shale oil. As an optional scheme, four key variables are selected as the constituent dimensions of the input features: the first is the mass ratio of kerogen to light oil molecules X1, with four levels: 5:1, 10:1, 15:1, and 20:1; the second is the pressure threshold of the semi-closed system X2, with five levels: 20 MPa, 25 MPa, 30 MPa, 35 MPa, and 40 MPa; the third is the final reaction temperature X3, with five levels: 2500 K, 3000 K, 3500 K, 4000 K, and 4500 K; and the fourth is the heating rate X4, with five levels: 10 K / ps, 15 K / ps, 20 K / ps, 25 K / ps, and 30 K / ps. Through the full combination design, a total of 4 × 5 × 5 × 5 = 500 different simulation conditions are generated.

[0133] For each operating condition, a complete simulation and parameter calculation process is executed independently, repeating all operations from steps 201 to 208. Specifically, for each operating condition, a corresponding molecular model is constructed according to its set mass ratio. Molecular dynamics simulation of the pyrolysis reaction in a semi-closed system is performed according to its set pressure threshold, final reaction temperature, and heating rate. Then, time-series parameters and product composition parameters are extracted from the simulation results. The simulated yield of each product and the simulated shale oil cracking rate are calculated. The simulated pyrolysis kinetic parameters (shale oil pyrolysis activation energy E and shale oil pyrolysis pre-exponential factor A) are obtained by fitting the first-order reaction kinetic equation with the Arrhenius equation. After completing the calculation for a single operating condition, the correspondence between a set of input features and output labels for that condition can be obtained.

[0134] To enhance the statistical robustness of the dataset and reduce the impact of random errors, five independent simulations were performed for each operating condition. Each simulation used the same input parameters but with slight differences in the initial molecular conformation. Taking the above operating conditions as an example, a total of 500 × 5 = 2500 sets of original samples were generated for 500 operating conditions.

[0135] After obtaining the original samples, quality control and outlier removal are required. Since the fitting quality of the pyrolysis kinetic parameters directly affects the reliability of the output labels, the coefficient of determination R0 for linear fitting is used. 2 As a screening criterion, for each sample group, check the R-value obtained in step 208. 2 Values, only retain R. 2 Samples with a value ≥0.95. In this embodiment, a total of R samples were removed. 2 There were 48 abnormal samples below the threshold, and 2452 valid samples were ultimately retained, forming the basis dataset for subsequent modeling.

[0136] At this point, each group of valid samples has input features and output labels. The input features are 4-dimensional vectors, including: X1 (mass ratio), X2 (pressure threshold, in MPa), X3 (final reaction temperature, in K), and X4 (heating rate, in K / ps).

[0137] The output labels are 7-dimensional vectors, including: Y1 (gaseous hydrocarbon yield, in %), Y2 (light oil yield, in %), Y3 (heavy oil yield, in %), Y4 (coke and bitumen yield, in %), Y5 (shale oil cracking rate, in %), Y6 (activation energy, in kJ / mol), and Y7 (pre-exponential factor, in s). -1 The formula for calculating the output label is as follows:

[0138] in, The simulated yield of gaseous hydrocarbons; The simulated yield of light oil; The simulated yield of heavy oil; The simulated yield of coke pitch; This represents the relative mass of gaseous hydrocarbons. This refers to the relative mass of light oils; M represents the relative mass of heavy oil; A represents the total relative mass of the initial molecular model; and A represents the pre-exponential factor of shale oil pyrolysis, in units of s. -1 E represents the activation energy of shale oil pyrolysis, in J / mol; R is the gas constant, with a value of 8.314 J / (mol). K); The slope of the fitted equation; Y1 represents the intercept of the fitted equation; Y7 represents the output labels.

[0139] The basic dataset is preprocessed with normalization to eliminate the influence of different physical dimensions on the training of the deep learning model, resulting in a multi-condition simulation sample set. The Min-Max normalization method is used to process each dimension of the input features and each dimension of the output labels independently. The normalization formula is:

[0140] Where X represents the original data of the sample; X min X is the minimum value of this dimension across all samples; max X is the maximum value of this dimension across all samples; norm The normalized data has its value range compressed to the [0,1] interval.

[0141] Normalization unifies features of different orders of magnitude and units to the same scale, which is beneficial to the convergence and training stability of neural network models.

[0142] This step, through systematic experimental design and data processing, constructed a multi-condition simulation sample set covering a wide parameter range, containing a large number of valid samples, and undergoing standardization. This sample set retains the physical realism of the microscopic simulation while possessing data standardization suitable for deep learning training, laying a solid data foundation for the subsequent construction of a high-precision multi-parameter coupled prediction model.

[0143] Step 210: Divide the multi-condition simulation sample set into a training sample set, a validation sample set, and a test sample set.

[0144] In this embodiment of the invention, the effective samples of the multi-condition simulation sample set (2452 effective samples in the above example) are randomly divided in a ratio of 7:1.5:1.5. 70% of the samples are allocated to the training sample set, i.e., 2452 × 70% ≈ 1716 groups. This data will be used for parameter updates during model training—the neural network gradually adjusts its weights and biases by repeatedly learning the mapping relationship between input features and output labels in the training samples, thereby continuously improving the model's predictive ability.

[0145] A validation set allocates 15% of the samples, approximately 368 groups (2452 × 15%). This data is not used for parameter updates during training but is instead used to monitor the model's training status. Specifically, after each training epoch, the current model makes predictions on the validation set and calculates the loss function value on the validation set. By observing the trend of the validation set loss, it's possible to determine whether the model is overfitting (i.e., the training set loss continues to decrease while the validation set loss begins to increase) or underfitting, allowing for timely adjustments to hyperparameters (such as learning rate, number of network layers, number of neurons, etc.) or triggering an early stopping mechanism. The existence of the validation set provides a basis for model tuning, avoiding the overfitting risk caused by relying solely on the training set for parameter tuning.

[0146] The test sample set allocates 15% of the samples, or 368 groups. This portion of data is completely isolated throughout the model training process; it is neither used for parameter updates nor for hyperparameter tuning, and is only used for final performance evaluation after model training is completed. By calculating various evaluation metrics on an unseen test set, the generalization prediction ability of the trained model for unknown operating conditions can be objectively measured, ensuring the authenticity and reliability of the model performance evaluation.

[0147] After the partitioning was completed, the statistical distribution of input features and output labels in the three subsets was analyzed, resulting in the statistical information table shown in Table 2: Table 2

[0148] Table 2 lists the minimum, maximum, mean, and standard deviation for each dimension. For example, the minimum value of the input feature X1 (mass ratio) is 5, the maximum value is 20, the mean is 12.3, and the standard deviation is 6.64; the minimum value of the output label Y1 (gaseous hydrocarbon yield) is 16.5%, the maximum value is 40.1%, the mean is 28.8%, and the standard deviation is 4.8%. This statistical information verifies the rationality of the data partitioning—the training set, validation set, and test set should have similar statistical distribution characteristics; it also provides a reference benchmark for subsequent model applications, such as using the minimum and maximum values ​​of each dimension when inversely normalizing predicted values.

[0149] This step establishes a standardized data foundation for the training, tuning, and evaluation of deep learning models through a scientific dataset partitioning strategy. The training set ensures that the model can fully learn the complex mapping relationship between input and output; the validation set provides a basis for hyperparameter optimization and overfitting monitoring; and the test set ensures the objectivity and credibility of model performance evaluation. These three elements complement each other, jointly supporting the construction of a high-precision, high-generalization-capability multi-parameter coupled prediction model.

[0150] Step 211: Using a preset adaptive optimizer and loss function, iteratively train the fully connected neural network and optimize the model parameters based on the training sample set. Then, based on the early stopping strategy, monitor the training process using the validation sample set to build an initial prediction model.

[0151] In this embodiment of the invention, Figure 6 This is a schematic diagram of a fully connected neural network architecture provided in an embodiment of the present invention, as shown below. Figure 6 As shown, the fully connected neural network adopts a three-level fully connected structure of "input layer-hidden layer-output layer". The number of neurons in the input layer is set to 4, which strictly corresponds to the dimensions of the input features (kerogen to light oil mass ratio X1, pressure threshold X2, final reaction temperature X3, heating rate X4). The hidden layer consists of two layers: the first hidden layer (hidden layer 1) has 90 neurons, and the second hidden layer (hidden layer 2) has 45 neurons. Both hidden layers use the ReLU activation function (formula f(x)=max(0,x)), which can effectively alleviate the gradient vanishing problem in deep network training, and at the same time introduce nonlinear transformation, enabling the network to fit complex multi-parameter coupling relationships. The number of neurons in the output layer is set to 7, which is consistent with the dimensions of the output labels (yields of each product Y1-Y4, shale oil cracking rate Y5, activation energy Y6, pre-exponential factor Y7), and the Linear activation function is used to directly output continuous numerical prediction results.

[0152] In this embodiment of the invention, the mean squared error (MSE) is selected as the loss function, which calculates the average of the squared deviations between the model's predicted values ​​and the true values ​​of the training samples. This function penalizes larger errors more strongly, which helps the model converge quickly. The optimizer uses the Adam adaptive optimizer, which is an optimization algorithm that combines momentum and adaptive learning rate. It can dynamically adjust the learning rate of each parameter based on the first and second moments of the gradient. In this embodiment, the initial learning rate η = 0.001, the momentum decay coefficient β1 = 0.9, and the adaptive learning rate decay coefficient β2 = 0.999 are set. These parameters ensure stable training and fast convergence.

[0153] Before training begins, the model weight matrix is ​​randomly initialized according to a normal distribution N(0,0.01), and the bias vector is initialized to 0. During training, the training sample set is input into the network in batches of 32. Each batch undergoes one forward propagation—the input features are sequentially processed by linear transformations and activation functions through each layer, ultimately yielding a 7-dimensional predicted value. Then, the error between the predicted and true values ​​is calculated using the loss function. Next, the gradient of the loss function with respect to the weights and biases of each layer is calculated using the backpropagation algorithm. Finally, the Adam optimizer is used to update the network parameters based on the gradient, completing the training of one batch. One iteration through all batches is called one training epoch.

[0154] To monitor for overfitting, the model performance is evaluated every 10 training epochs using a validation set, and the loss value on the validation set is calculated. An early stopping strategy is also introduced: when the validation set loss no longer decreases for 8 consecutive epochs, the model is considered to have converged or begun to overfit, training is immediately stopped, and the model parameters at the point of lowest validation set loss are saved as the final model. This strategy effectively avoids ineffective long-term training and the risk of overfitting.

[0155] In this embodiment of the invention, the maximum number of iterations is set to 800 rounds. Figure 7 This is a schematic diagram of the loss value during actual training, provided as an embodiment of the present invention. Figure 7 As shown, the horizontal axis represents the number of iterations, ranging from 0 to 400 with an interval of 50; the vertical axis represents the training loss (MSE), ranging from 0.00 to 0.10 with an interval of 0.02; the red curve represents the training set loss, and the blue curve represents the validation set loss. The loss function used for both is MSE. Figure 7 As shown, the loss decreases rapidly in the first 100 rounds, the rate of decrease slows down from 100 to 300 rounds, the validation set loss stabilizes after 350 rounds, and finally early stopping is triggered at 380 rounds, ending the training.

[0156] In this embodiment of the invention, to further enhance the model's generalization ability and prevent overfitting, Dropout layers are added after each of the two hidden layers, with a dropout probability set to 0.1. This means that in each training iteration, 10% of the outputs of the hidden layer neurons are randomly set to zero, forcing the network to learn more robust feature representations and avoiding excessive reliance on local features.

[0157] This step achieves accurate learning of the complex nonlinear mapping from input features to output labels through a fully connected neural network architecture, optimization algorithms, and training strategies. The early stopping strategy and validation set monitoring mechanism ensure that the model has good generalization performance, and Dropout regularization further improves the robustness of the model. The initial prediction model obtained by the final training lays a reliable model foundation for subsequent rapid prediction and process optimization.

[0158] Step 212: Perform performance testing on the initial prediction model using the test sample set. If the test passes, proceed to step 213; if the test fails, proceed to step 210.

[0159] In this embodiment of the invention, the test sample set consists of 368 groups of samples that were not involved in training and validation, as defined in step 210. These samples are completely isolated throughout the training process, ensuring the objectivity of the evaluation results. Performance evaluation uses three metrics: Mean Absolute Error (MAE), Coefficient of Determination (R²), and Fibre Channel Index (FCL). 2 The mean absolute error (MAE) measures the average deviation between predicted and actual values; the coefficient of determination (R²) is used to measure the difference between predicted and actual values.2 The _________ reflects the goodness of fit of the model to the data, with a value ranging from 0 to 1. The closer to 1, the stronger the model's explanatory power. The maximum relative error (MRE) measures the maximum percentage deviation between the predicted and actual values. The formulas for each indicator are as follows:

[0160] Where W is the number of samples in the test sample set (368); h is the dimension of the output label (7). This represents the true value of the j-th label for the i-th sample. The corresponding model prediction value; Let be the average value of the j-th label across all test samples.

[0161] Specifically, if all performance metrics of the test sample set meet the performance requirements, the test is passed, and step 213 is executed; if at least one performance metric of the test sample set fails to meet the performance requirements, the test fails, and step 210 is executed to adjust the dataset partition or retrain the hyperparameters.

[0162] As an optional solution, the preset performance requirements are: MAE ≤ 0.005, R 2 ≥0.97, MRE≤3%.

[0163] The 368 test samples were input into the trained initial prediction model to obtain a 7-dimensional predicted value for each sample. The predicted value was then compared with the true value to calculate the above indicators. The results are shown in Table 3. Table 3

[0164] As shown in Table 3, the MAE is 0.0027, which is far below the threshold of 0.005; R 2 A mean squared error of 0.99 indicates that the model can explain 99% of the data variation; the mean recurrence interval (MRE) is 2.2%, which is also less than the 3% limit. All indicators meet the preset requirements, indicating that the initial prediction model has high prediction accuracy and good generalization ability.

[0165] To more intuitively demonstrate the model's predictive performance, a set of known operating conditions (kerogen to light oil mass ratio 10:1, pressure threshold 30 MPa, final reaction temperature 3500 K, heating rate 20 K / ps) was randomly selected from the test set. The input features were normalized and then input into the model to obtain normalized predicted values. These were then inversely normalized to restore the actual physical quantities, and the comparison with the actual values ​​is shown in Table 4. Table 4

[0166] As shown in Table 4, the relative prediction error of the initial prediction model for each output label is within 5%, and the relative prediction error of the core indicator shale oil cracking rate is only 1.39%, which further verifies the reliability of the model prediction.

[0167] Furthermore, to verify the rationality of the input feature selection, this embodiment of the invention compared the performance of a 2D input model containing only mass ratio and pressure threshold with that containing all 4D features on the test set. The results are shown in Table 5: Table 5

[0168] As can be seen from the comparison, after adding the final reaction temperature and heating rate as input features, the overall prediction accuracy of the model is significantly improved. In particular, the maximum relative error of the shale oil cracking rate prediction is reduced from 4.85% to 0.85%, which fully demonstrates the scientific nature and effectiveness of the 4-dimensional input features.

[0169] Step 213: Determine the initial prediction model as a multi-parameter coupled prediction model.

[0170] In this embodiment of the invention, after verification with a test sample set, the prediction accuracy and generalization ability of the initial prediction model both meet the preset standards, and the initial prediction model is determined to be a multi-parameter coupled prediction model. The multi-parameter coupled prediction model encapsulates the network weights, biases, and architecture parameters trained in step 211, and can be directly used for rapid prediction of unknown working conditions.

[0171] Furthermore, based on the trained multi-parameter coupled prediction model, multi-parameter coupled optimization is carried out. The specific process includes: The optimization objective is set as follows: maximize the shale oil cracking rate (Y5→max); the variable constraints are: kerogen to light oil mass ratio X1∈[5:1, 20:1], pressure threshold X2∈[20MPa, 40MPa], final reaction temperature X3∈[2500K, 4500K], and heating rate X4∈[10K / ps, 30K / ps].

[0172] Optimization Algorithm: Particle Swarm Optimization (PSO) is used, with the multi-parameter coupled prediction model as the fitness function to search for the optimal parameter combination.

[0173] Optimization results: The optimal parameter combination was obtained as X1=10:1, X2=35MPa, X3=4000K, X4=20K / ps. Under this condition, the shale oil cracking rate was 45.7%, which was 14.5 percentage points higher than the lowest cracking rate (31.2%).

[0174] Step 214: Normalize the operating parameters to be predicted to obtain the input features to be predicted.

[0175] In this embodiment of the invention, when it is necessary to predict a new working condition in a practical application, the working condition parameters to be predicted are obtained, namely: 4-dimensional input features, including: kerogen to light oil mass ratio X1, pressure threshold X2, final reaction temperature X3, and heating rate X4.

[0176] Since the input features are all normalized using Min-Max during model training, the original parameters to be predicted must be transformed using the same normalization rule to ensure that the distribution of the input data is consistent with that of the model training data. The normalization formula is described in step 209 and will not be repeated here. By using the normalization formula, the operating parameters to be predicted are mapped to the [0,1] interval, resulting in the input feature vector to be predicted, which can then be used in subsequent steps.

[0177] This step ensures that the model input and training data are on the same scale, eliminating the influence of dimensions and laying the foundation for accurate predictions later.

[0178] Step 215: Input the input features to be predicted into the multi-parameter coupled prediction model for forward propagation to generate normalized prediction output features.

[0179] In this embodiment of the invention, a file of a multi-parameter coupled prediction model is loaded. This file contains the trained neural network weights, biases, and network architecture parameters. The input features to be predicted are input into the model, and forward propagation calculation is performed. Forward propagation is the forward computation process of the neural network, and the forward propagation process is as follows:

[0180] Where X is; , , b1, b2, and b3 are the weight matrices from the input layer to the first hidden layer, from the first hidden layer to the second hidden layer, and from the second hidden layer to the output layer, with dimensions (4, 90), (90, 45), and (45, 7), respectively; b1, b2, and b3 are the bias vectors of the corresponding layers, with dimensions (90,), (45,), and (7,), respectively; X is the input feature vector; H1 and H2 are the outputs of the two hidden layers, respectively. These are the normalized predicted output features.

[0181] It is worth noting that (90,) represents a one-dimensional vector of length 90; (45,) represents a one-dimensional vector of length 45; and (7,) represents a one-dimensional vector of length 7.

[0182] The entire forward propagation process is a single forward calculation, without involving backpropagation or parameter updates, thus resulting in extremely fast computation speed. This process eliminates the need for repeated, time-consuming molecular dynamics simulations or dynamic parameter fitting, achieving an instantaneous response from input conditions to output predictions.

[0183] In this embodiment of the invention, the normalized predicted output feature is a 7-dimensional vector, with each component in the interval [0,1], corresponding to the normalized values ​​of gaseous hydrocarbon yield, light oil yield, heavy oil yield, coke pitch yield, shale oil cracking rate, activation energy, and pre-exponential factor, respectively.

[0184] This step, by encapsulating the trained deep learning model, enables rapid prediction of unknown operating conditions. It fundamentally solves the pain point that traditional molecular dynamics simulations for a single operating condition take several days and cannot meet the needs of rapid engineering iteration, providing efficient technical support for real-time optimization and decision-making of in-situ conversion processes.

[0185] Step 216: Perform inverse normalization on the normalized predicted output features to generate the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

[0186] In this embodiment of the invention, the normalized values ​​lack direct physical meaning and cannot be directly used for engineering analysis and decision-making. Therefore, they are restored to true predicted values ​​with actual physical units through inverse normalization. The target yields of each product include gaseous hydrocarbon yield, light oil yield, heavy oil yield, and coke pitch yield. The target pyrolysis kinetic parameters include shale oil pyrolysis activation energy and shale oil pyrolysis pre-exponential factor.

[0187] Specifically, through the inverse normalization formula Y re-pred =Y pred ×(Y max Y min )+Y min , where Y re-pred The true predicted values ​​after inverse normalization are: the target yield of each product, the target shale oil cracking rate, and the target pyrolysis kinetic parameters; Y pred Y represents the normalized predicted output features. max Y represents the maximum value of the output label dimension in the training set. min This is the minimum value of the output label dimension in the training set.

[0188] This step, through inverse normalization, restores the abstract numerical values ​​output by the model to engineering parameters with clear physical meaning, realizing a complete data closed loop from microscopic simulation to macroscopic application. The prediction results after inverse normalization maintain the same dimensions and order of magnitude as the original experimental data, ensuring the interpretability and engineering applicability of the prediction results, and providing direct and usable quantitative basis for parameter optimization, scheme comparison, and real-time decision-making in in-situ conversion processes.

[0189] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0190] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0191] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making result; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0192] The technical solution of the multi-parameter coupled prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil provided in this invention embodiment is based on various medium- and low-maturity shale oil samples collected by pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil is constructed based on kerogen molecules and selected light oil molecules; based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, the molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model is performed to obtain a multi-condition parameter set, which includes time-series parameters and product composition parameters; the simulated yield of each product and the simulated shale oil cracking rate are calculated based on the product composition parameters, and the simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius formula, thus constructing a multi-condition simulation sample set; By using a large number of simulated operating conditions as a sample set, a model is constructed on a pre-defined fully connected neural network to generate a multi-parameter coupled prediction model. This model predicts parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product. By constructing a semi-closed, dynamically pressure-controlled molecular simulation system, accurate replication of the actual formation environment is achieved, significantly improving the simulation realism and reliability of the in-situ conversion process. Furthermore, by coupling reaction kinetic equations with a deep learning model, rapid calculation and multi-parameter coupled prediction of pyrolysis kinetic parameters are realized, greatly improving parameter acquisition efficiency and prediction accuracy. This allows for comprehensive consideration of the coupling effects of key process parameters, providing efficient and scientific technical support for the optimized design of shale oil in-situ conversion processes.

[0193] Figure 8 This is a schematic diagram of a multi-parameter coupled prediction device for semi-closed in-situ transformation of medium- and low-maturity shale oil, provided in an embodiment of the present invention. This device is used to execute the aforementioned multi-parameter coupled prediction method for semi-closed in-situ transformation of medium- and low-maturity shale oil. Figure 8 As shown, the device includes: a molecular model construction module 51, a dynamic simulation execution module 52, a sample set construction module 53, a deep learning training module 54, and a parameter prediction module 55.

[0194] The molecular model construction module 51 is used to collect various medium- and low-maturity shale oil samples based on thermal pyrolysis gas chromatography-mass spectrometry analysis, and to construct a three-dimensional periodic molecular model of the target shale oil based on kerogen molecules and selected light oil molecules.

[0195] The dynamic simulation execution module 52 is used to perform molecular dynamics simulation of the deheating reaction of the target shale oil three-dimensional periodic molecular model based on the preset boundary conditions, simulation parameters and selected reaction force field of the semi-closed system, and obtain a multi-condition parameter set, which includes time parameters and product composition parameters.

[0196] The sample set construction module 53 is used to calculate the simulated yield and simulated shale oil cracking rate of each product based on the product composition parameters, and to couple the first-order reaction kinetic equation with the Arrhenius formula to obtain the simulated pyrolysis kinetic parameters, and construct a multi-condition simulation sample set.

[0197] The deep learning training module 54 is used to build a model for a pre-defined fully connected neural network using a multi-condition simulation sample set, and generate a multi-parameter coupled prediction model.

[0198] The parameter prediction module 55 is used to perform parameter prediction based on the operating condition parameters to be predicted through a multi-parameter coupled prediction model, and generate the target yield, target shale oil cracking rate and target pyrolysis kinetic parameters of each product.

[0199] In this embodiment of the invention, the molecular model construction module 51 includes a component analysis unit 511, a mass ratio range generation unit 512, a mixing unit 513, and an optimized annealing unit 514.

[0200] The component analysis unit 511 is used to perform thermal pyrolysis gas chromatography-mass spectrometry analysis on various medium- and low-maturity shale oil samples to obtain hydrocarbon product composition data of different medium- and low-maturity shale oil samples.

[0201] The mass ratio range generation unit 512 is used to generate a simulated mass ratio range based on the component data of kerogen molecules and selected light oil molecules in the hydrocarbon product component composition data.

[0202] The mixing unit 513 is used to mix kerogen molecules and light oil molecules according to different simulated mass ratios to construct initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios.

[0203] The optimized annealing unit 514 is used to perform structural optimization and annealing on the initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios, and to generate the target three-dimensional periodic molecular model of shale oil.

[0204] In this embodiment of the invention, the dynamic simulation execution module 52 includes a simulation unit 521 and a monitoring unit 522.

[0205] Simulation unit 521 is used to perform multiple semi-closed system deheating reaction molecular dynamics simulations on the target shale oil three-dimensional periodic molecular model according to different simulation parameters and using a reaction force field. It controls the escape of gaseous hydrocarbon molecules according to boundary response rules. When the system pressure is greater than or equal to the pressure threshold, it controls the system pressure to maintain dynamic equilibrium.

[0206] The monitoring unit 522 is used to monitor and record the timing parameters and product composition parameters during the simulation process, and generate a multi-condition parameter set.

[0207] In this embodiment of the invention, the sample set construction module 53 includes a coupling unit 531 and a construction unit 532.

[0208] The coupling unit 531 is used to couple the first-order reaction kinetic equation with the Arrhenius equation, and calculate the simulated pyrolysis kinetic parameters based on the simulated shale oil cracking rate and time series parameters.

[0209] The construction unit 532 is used to construct a multi-condition simulation sample set based on the simulated yield of each product, the simulated shale oil cracking rate, and the simulated pyrolysis kinetic parameters under different working conditions.

[0210] In this embodiment of the invention, the deep learning training module 54 includes a partitioning unit 541, a training unit 542, and a testing unit 543.

[0211] The partitioning unit 541 is used to divide the multi-condition simulation sample set into a training sample set, a validation sample set, and a test sample set.

[0212] The training unit 542 is used to iteratively train the fully connected neural network and optimize the model parameters based on the training sample set using a preset adaptive optimizer and loss function, and to monitor the training process using a validation sample set based on an early stopping strategy, thereby building an initial prediction model.

[0213] The testing unit 543 is used to perform performance testing on the initial prediction model using a test sample set. If the test passes, the initial prediction model is determined to be a multi-parameter coupled prediction model. If the test fails, the partitioning unit 541 is triggered to repeat the steps of partitioning the multi-condition simulation sample set into a training sample set, a validation sample set, and a test sample set.

[0214] In this embodiment of the invention, the parameter prediction module 55 includes: a normalization unit 551, a forward propagation unit 552, and an anti-normalization unit 553.

[0215] The normalization unit 551 is used to normalize the operating parameters to be predicted, so as to obtain the input features to be predicted.

[0216] The forward propagation unit 552 is used to input the input features to be predicted into the multi-parameter coupled prediction model for forward propagation, and generate normalized prediction output features.

[0217] The inverse normalization unit 553 is used to inverse normalize the predicted output features after normalization to generate the target yield, target shale oil cracking rate and target pyrolysis kinetic parameters for each product.

[0218] In this embodiment of the invention, multiple medium- and low-maturity shale oil samples are collected based on pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil is constructed based on kerogen molecules and selected light oil molecules. Based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, the molecular dynamics simulation of the depyrolysis reaction of the target shale oil is performed on the three-dimensional periodic molecular model, resulting in a multi-condition parameter set, which includes time-series parameters and product composition parameters. The simulated yield and simulated shale oil cracking rate of each product are calculated based on the product composition parameters, and the simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius equation, thus constructing a multi-condition simulation sample set. Using the multi-condition simulation sample set, the preset... A fully connected neural network is used to construct the model, generating a multi-parameter coupled prediction model. This model predicts parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product. By constructing a semi-closed, dynamically pressure-controlled molecular simulation system, accurate replication of the actual formation environment is achieved, significantly improving the simulation realism and reliability of the in-situ conversion process. Furthermore, by coupling reaction kinetic equations with a deep learning model, rapid calculation and multi-parameter coupled prediction of pyrolysis kinetic parameters are realized, greatly improving parameter acquisition efficiency and prediction accuracy. This approach comprehensively considers the coupling effects of key process parameters, providing efficient and scientific technical support for the optimized design of shale oil in-situ conversion processes.

[0219] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0220] This invention provides a computer device, including a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described embodiment of the multi-parameter coupling prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil. For a detailed description, please refer to the above-described embodiment of the multi-parameter coupling prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil.

[0221] The following is for reference. Figure 9 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0222] like Figure 9As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0223] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed in storage section 608 as needed.

[0224] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0225] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0226] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0227] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0228] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0229] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0230] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0231] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0232] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0233] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0234] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0235] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0236] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A multi-parameter coupled prediction method for semi-closed in-situ transformation of medium- and low-maturity shale oil, characterized in that, The method includes: Multiple medium- and low-maturity shale oil samples were collected based on thermal pyrolysis gas chromatography-mass spectrometry analysis, and a three-dimensional periodic molecular model of the target shale oil was constructed based on kerogen molecules and selected light oil molecules. Based on the pre-set boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, the molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model is carried out to obtain a multi-condition parameter set, which includes time-series parameters and product composition parameters. The simulation parameters include: heating rate, initial temperature, and final reaction temperature; the boundary conditions of the semi-closed system include: boundary response parameters and boundary response rules; the boundary response parameters include: pressure threshold; the boundary response rules are: when the system pressure is greater than or equal to the preset pressure threshold, gaseous hydrocarbon molecules with a carbon atom number within a preset range are allowed to escape through the boundary; when the system pressure is less than the pressure threshold, hydrocarbon molecules are not allowed to escape. The simulated yield and simulated shale oil cracking rate of each product are calculated based on the product composition parameters. The simulated pyrolysis kinetic parameters are obtained by fitting the first-order reaction kinetic equation with the Arrhenius formula and constructing a multi-condition simulation sample set. Using the multi-condition simulation sample set, a model is constructed on the preset fully connected neural network to generate a multi-parameter coupled prediction model; The multi-parameter coupled prediction model is used to predict parameters based on the operating conditions to be predicted, thereby generating the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

2. The method for semi-closed in-situ conversion multi-parameter coupled prediction of medium- and low-maturity shale oil according to claim 1, characterized in that, The method involves collecting various medium- and low-maturity shale oil samples based on pyrolysis gas chromatography-mass spectrometry analysis, and constructing a three-dimensional periodic molecular model of the target shale oil based on kerogen molecules and selected light oil molecules, including: The various medium- and low-maturity shale oil samples were subjected to pyrolysis gas chromatography-mass spectrometry analysis to obtain hydrocarbon product composition data for different medium- and low-maturity shale oil samples. Based on the component data of kerogen molecules and selected light oil molecules in the hydrocarbon product composition data, a simulated mass ratio range is generated. Kerogen molecules and light oil molecules were mixed separately according to different simulated mass ratios to construct initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios. The initial three-dimensional periodic molecular models of shale oil with different simulated mass ratios were structurally optimized and annealed to generate the target three-dimensional periodic molecular model of shale oil.

3. The method for semi-closed in-situ conversion multi-parameter coupled prediction of medium- and low-maturity shale oil according to claim 1, characterized in that, Based on the preset boundary conditions, simulation parameters, and selected reaction force field of the semi-closed system, a molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model is performed to obtain a multi-condition parameter set, including: According to different simulation parameters, the reaction force field is used to perform multiple semi-closed system deheating reaction molecular dynamics simulations on the target shale oil three-dimensional periodic molecular model, and the gaseous hydrocarbon molecules are controlled to escape according to the boundary response rules; when the system pressure is greater than or equal to the pressure threshold, the system pressure is controlled to maintain dynamic equilibrium. The timing parameters and product composition parameters during the simulation process are monitored and recorded to generate the multi-condition parameter set.

4. The method for multi-parameter coupled prediction of semi-closed in-situ transformation of medium- and low-maturity shale oil according to claim 1, characterized in that, The coupled first-order reaction kinetic equations are fitted with the Arrhenius formula to obtain simulated pyrolysis kinetic parameters, and a multi-condition simulation sample set is constructed, including: The first-order reaction kinetic equation was coupled with the Arrhenius equation, and the simulated pyrolysis kinetic parameters were calculated based on the simulated shale oil cracking rate and time series parameters. Based on the simulated yield of each product, the simulated shale oil cracking rate, and the simulated pyrolysis kinetic parameters under different operating conditions, a multi-condition simulation sample set was constructed.

5. The method for multi-parameter coupled prediction of semi-closed in-situ transformation of medium- and low-maturity shale oil according to claim 1, characterized in that, The step of constructing a model for a pre-defined fully connected neural network using the multi-condition simulation sample set to generate a multi-parameter coupled prediction model includes: The multi-condition simulation sample set is divided into a training sample set, a validation sample set, and a test sample set. The fully connected neural network is iteratively trained and its model parameters are optimized using a preset adaptive optimizer and loss function based on the training sample set. The training process is monitored using a validation sample set based on an early stopping strategy to build an initial prediction model. The performance of the initial prediction model is tested using the test sample set. If the test passes, the initial prediction model will be determined as a multi-parameter coupled prediction model; If the test fails, repeat the step of dividing the multi-condition simulation sample set into a training sample set, a verification sample set, and a test sample set.

6. The method for multi-parameter coupled prediction of semi-closed in-situ transformation of medium- and low-maturity shale oil according to claim 1, characterized in that, The process involves using the multi-parameter coupled prediction model to predict parameters based on the operating conditions to be predicted, generating target yields, target shale oil cracking rates, and target pyrolysis kinetic parameters for each product, including: The operating parameters to be predicted are normalized to obtain the input features to be predicted. The input features to be predicted are input into a multi-parameter coupled prediction model for forward propagation to generate normalized prediction output features. The normalized predicted output features are then denormalized to generate the target yield, target shale oil cracking rate, and target pyrolysis kinetic parameters for each product.

7. A semi-closed in-situ multi-parameter coupled prediction device for medium- and low-maturity shale oil conversion, characterized in that, The device includes: The molecular model construction module is used to construct a three-dimensional periodic molecular model of the target shale oil based on various medium- and low-maturity shale oil samples collected by thermal pyrolysis gas chromatography-mass spectrometry analysis, and based on kerogen molecules and selected light oil molecules. The dynamic simulation execution module is used to perform molecular dynamics simulation of the dethermation reaction of the target shale oil three-dimensional periodic molecular model based on the preset boundary conditions, simulation parameters and selected reaction force field of the semi-closed system, and to obtain a multi-condition parameter set, which includes time parameters and product composition parameters. The simulation parameters include: heating rate, initial temperature, and final reaction temperature; the boundary conditions of the semi-closed system include: boundary response parameters and boundary response rules; the boundary response parameters include: pressure threshold; the boundary response rules are: when the system pressure is greater than or equal to the preset pressure threshold, gaseous hydrocarbon molecules with a carbon atom number within a preset range are allowed to escape through the boundary; when the system pressure is less than the pressure threshold, hydrocarbon molecules are not allowed to escape. The sample set construction module is used to calculate the simulated yield and simulated shale oil cracking rate of each product based on the product composition parameters, and to couple the first-order reaction kinetic equation with the Arrhenius formula to obtain the simulated pyrolysis kinetic parameters and construct a multi-condition simulation sample set. The deep learning training module is used to build a model of a preset fully connected neural network using the multi-condition simulation sample set, and generate a multi-parameter coupled prediction model. The parameter prediction module is used to predict parameters based on the operating parameters to be predicted through the multi-parameter coupled prediction model, and generate the target yield, target shale oil cracking rate and target pyrolysis kinetic parameters of each product.

8. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the multi-parameter coupled prediction method for semi-closed in-situ transformation of medium- and low-maturity shale oil as described in any one of claims 1 to 6.

9. A computer device comprising a memory and a processor, the memory for storing information including program instructions, and the processor for controlling the execution of the program instructions, characterized in that, When the program instructions are loaded and executed by the processor, the method for multi-parameter coupling prediction of semi-closed in-situ conversion of medium- and low-maturity shale oil as described in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the multi-parameter coupling prediction method for semi-closed in-situ conversion of medium- and low-maturity shale oil as described in any one of claims 1 to 6.

Images