Physical information constrained petrochemical material mixture modeling physical property calculation system and method

Abstract

The present invention relates to the field of petrochemical material physical property calculation, and specifically to a physical information constrained petrochemical material mixed modeling physical property calculation system and method, comprising: a data processing module to realize the standardized cleaning and fusion of multi-source heterogeneous data; a multi-scale feature module to realize material property characterization through quantum-mesoscopic-macroscopic cross-scale modeling; a physical constraint module to embed thermodynamic constitutive equations and phase equilibrium criteria to ensure model self-consistency; a component interaction module to construct a non-ideal mixing effect prediction model based on deep potential energy field theory and establish a component interaction knowledge graph; an uncertainty quantification module to evaluate the prediction confidence interval using a Bayesian deep learning method, systematically integrating physical mechanisms and data-driven methods to achieve a 15% to 20% reduction in physical property prediction errors, and at the same time providing a visual analysis tool for molecular structure-physical property association to support process optimization and new product development decisions, meeting the petrochemical industry's core demand for high-precision and explainable physical property predictions.

Description

Technical Field

[0001] The present invention relates to the field of petrochemical material physical property calculation, and specifically to a physical information constrained petrochemical material mixture modeling physical property calculation system and method thereof, which can accurately predict various physical and chemical properties of petrochemical material mixtures based on multi-scale feature characterization and physical constraints. Background Art

[0002] Predicting the physical properties of petrochemical materials is a critical step in the petrochemical industry, crucial for process design, product development, and quality control. Traditional physical property prediction methods fall into three main categories: experimental measurement, empirical correlation, and theoretical calculation. While experimental measurement offers high accuracy, it is time-consuming and costly, making it difficult to meet the demands of rapid R&D. Empirical correlation relies on extensive historical data, limiting its predictive power. Theoretical calculations can predict physical properties based on molecular structure, but their computational complexity makes them difficult to apply to complex mixtures.

[0003] In recent years, with the advancement of machine learning technology, data-driven physical property prediction methods, such as QSPR (quantitative structure-property relationship) and artificial neural networks, have gradually emerged. These methods can learn complex nonlinear relationships from large amounts of data, enabling relatively rapid property prediction. However, existing data-driven methods suffer from the following problems: First, most are single-component models, making it difficult to accurately account for component interactions in mixtures; second, predictions often violate fundamental physical laws, such as thermodynamics; third, feature representation is limited to a single scale, making it difficult to fully capture the relationship between molecular structure and macroscopic properties; fourth, prediction capabilities are weak in areas with sparse data; and fifth, uncertainty assessment is insufficient, making it difficult to support reliable decision making.

[0004] Therefore, there is an urgent need for a petrochemical material mixture physical property prediction system that can integrate physical laws and multi-scale characteristics to improve prediction accuracy and reliability while ensuring physical rationality. Summary of the Invention

[0005] The purpose of the invention is to provide a physical information constrained petrochemical material mixture modeling physical property calculation system and method thereof, which integrates physical law constraints and multi-scale feature characterization, can accurately predict various physical and chemical properties of petrochemical material mixtures, and provide reliable uncertainty assessment.

[0006] The present invention proposes a physical information constrained petrochemical material mixture modeling and physical property calculation system, comprising:

[0007] A data processing and management module is configured to: collect multi-source data of petrochemical materials; clean and quality control the multi-source data; and integrate the multi-source data into standard format data; a multi-scale feature characterization module is communicatively connected to the data processing and management module and is configured to: receive the standard format data; construct a multi-level petrochemical material feature characterization from quantum scale to macro scale; and achieve the fusion of features at different scales; a physical constraint modeling module is communicatively connected to the multi-scale feature characterization module and is configured to: receive the multi-level petrochemical material feature characterization; embed thermodynamic laws and physical law constraints into the model structure; and ensure that the physical property prediction results meet the thermodynamic consistency requirements; a component interaction modeling module is communicatively connected to the multi-scale feature characterization module and the physical constraint modeling module and is configured to: accurately capture the complex interactions between different components in a petrochemical material mixture; predict non-ideal mixing effects between components; and construct a component interaction knowledge base;

[0008] An uncertainty quantification module is communicatively connected to the physical constraint modeling module and the component interaction modeling module, and is configured to:

[0009] Evaluate the reliability of physical property prediction results; quantify the sources and magnitude of prediction uncertainty; generate confidence intervals for prediction results; the physical property prediction and analysis module is communicated with the physical constraint modeling module, the component interaction modeling module and the uncertainty quantification module, and is used to: predict various physical and chemical properties of petrochemical materials based on the outputs of the above modules; analyze the relationship between molecular structure and physical properties; and provide visualization and interpretation functions.

[0010] A physical information-constrained method for calculating physical properties of petrochemical material mixture modeling includes: collecting multi-source data of petrochemical materials, and cleaning, quality controlling, and standardizing the multi-source data to obtain data in a standard format; based on the standard format data, constructing a multi-level petrochemical material feature representation from the quantum scale to the macroscopic scale, and realizing the fusion of features at different scales; embedding thermodynamic laws and physical law constraints into the model structure to ensure that the physical property prediction results meet the thermodynamic consistency requirements; accurately capturing the complex interactions between different components in the petrochemical material mixture, predicting the non-ideal mixing effects between components, and constructing a component interaction knowledge base; evaluating the reliability of the physical property prediction results, quantifying the sources and size of the uncertainty in the prediction, and generating confidence intervals for the prediction results; based on the above steps, predicting various physical and chemical properties of petrochemical materials, analyzing the relationship between molecular structure and physical properties, and providing visualization and interpretation functions.

[0011] The beneficial effects of the present invention include:

[0012] 1. By integrating multi-scale features, we comprehensively characterize the structure-property relationships of petrochemical materials from the quantum to the macroscale, improving prediction accuracy. This improves property prediction accuracy by over 30% compared to traditional methods, especially for complex mixture systems.

[0013] 2. By embedding thermodynamic laws and physical constraints into the model structure, the prediction results are ensured to conform to basic physical laws, avoiding non-physical interpretations. The thermodynamic consistency error of the system's prediction results is reduced to less than one-fifth that of traditional methods.

[0014] 3. A specially designed component interaction modeling module accurately captures the complex interactions and synergistic effects between different components in a mixture, overcoming the limitations of single-component methods. For highly non-ideal mixture systems, the activity coefficient prediction error is reduced by over 40%.

[0015] 4. Ability to simultaneously predict more than 15 key physical properties, including thermodynamic properties, transport properties, combustion properties and environmental safety properties, to meet the petrochemical industry's comprehensive needs for multiple physical property parameters.

[0016] 5. Through systematic uncertainty quantification techniques, we provide reliable forecast uncertainty assessments to support risk-based decision making. The confidence interval coverage of the forecast results reaches over 95%, which is highly consistent with the actual statistical distribution.

[0017] 6. Through transfer learning and knowledge distillation techniques, we effectively address data sparsity, achieve efficient model generalization, and reduce reliance on large amounts of data. With only 10% training data, model performance drops by no more than 15%.

[0018] 7. Provide clear explanations of physical meaning and intuitive visualization functions to assist scientific understanding and process optimization, and enhance user trust.

[0019] 8. Seamless integration with existing petrochemical industry process simulation systems and process design platforms facilitates practical application and promotion. The system can be integrated with mainstream simulation software such as Aspen Plus and HYSYS, and supports standard data format exchange. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a diagram of the overall system architecture of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific examples. It should be understood by those skilled in the art that these examples are only used to illustrate the present invention and should not be regarded as limiting the present invention.

[0022] Reference Figure 1The physical information-constrained petrochemical material mixture modeling and physical property calculation system provided by the present invention includes: a data processing and management module 1, a multi-scale feature characterization module 2, a physical constraint modeling module 3, a component interaction modeling module 4, an uncertainty quantification module 5, and a physical property prediction and analysis module 6. Furthermore, the system may also include a model training and optimization module 7 and a system integration and application module 8.

[0023] The Data Processing and Management Module 1 collects multi-source data on petrochemical materials and cleans, quality-controls, and standardizes this data to generate standardized data. The Multiscale Characterization Module 2 receives this standardized data and constructs a multi-level petrochemical material characterization from the quantum to the macroscale, integrating features at different scales. The Physical Constraint Modeling Module 3 receives this multi-level petrochemical material characterization and embeds thermodynamic laws and physical constraints into the model structure to ensure that physical property predictions meet thermodynamic consistency requirements. The Component Interaction Modeling Module 4 accurately captures the complex interactions between different components in petrochemical material mixtures, predicts non-ideal mixing effects between components, and constructs a component interaction knowledge base. The Uncertainty Quantification Module 5 assesses the reliability of physical property predictions, quantifies the sources and magnitude of uncertainty in the predictions, and generates confidence intervals for the predictions. Based on the outputs of the above modules, the Physical Property Prediction and Analysis Module 6 predicts various physical and chemical properties of petrochemical materials, analyzes the relationship between molecular structure and physical properties, and provides visualization and interpretation capabilities.

[0024] Preferably, the system also includes a model training and optimization module 7 for implementing multi-task joint learning, optimizing physical constraint enforcement, optimizing model architecture and hyperparameters, evaluating model performance, and selecting the optimal model. Furthermore, the system may also include a system integration and application module 8 for providing a user interface, integrating with industrial systems, performing data exchange and standardization, supporting system deployment and expansion, and adapting to different application scenarios.

[0025] The data processing and management module 1 includes a multi-source data acquisition unit, a data cleaning and quality control unit, a data integration and standardization unit, and a knowledge graph construction unit.

[0026] The multi-source data acquisition unit is used to acquire petrochemical material data from a variety of sources. In one embodiment of the present invention, the multi-source data acquisition unit can collect data from three main data sources: experimental databases (such as the NIST ThermoDataEngine and the DIPPR database), theoretical calculation results (such as quantum chemistry calculations and molecular dynamics simulations), and industrial process data (such as factory production data and quality control data). Preferably, the multi-source data acquisition unit supports multiple data formats, including CSV, Excel, JSON, and XML, and provides an API interface for automated data collection.

[0027] The Data Cleaning and Quality Control Unit communicates with the Multi-Source Data Acquisition Unit, receives the collected raw data, and performs cleaning and quality control on it. This unit first detects and processes outliers, using statistical methods (such as the Z-score and IQR) and physical laws (such as the laws of thermodynamics) to identify abnormal data. For example, for liquid density data, density typically decreases as temperature increases. Data points where density increases with increasing temperature are likely outliers. Second, the unit verifies the thermodynamic consistency of the data, checking whether the data satisfies fundamental thermodynamic equations such as Maxwell's relation. Finally, the data is scored for quality, evaluating it based on three dimensions: completeness, accuracy, and consistency. Weighting is assigned based on factors such as the reliability of the data source, data acquisition methods, and data age. For example, NIST-certified standard reference data may receive a quality score of 0.9-1.0, while non-peer-reviewed literature data may only receive a score of 0.5-0.7.

[0028] The data integration and standardization unit communicates with the data cleaning and quality control unit, receives the cleaned data, and integrates it into a standard format. This unit first unifies the physical property data format and defines a standard structure that includes substance identifiers, condition parameters (temperature, pressure, etc.), and physical property values. Secondly, it standardizes molecular representations and converts molecular structures in different formats into SMILES or InChI format. For example, molecular structure diagrams, molecular names, or CAS numbers are converted into standard SMILES representations. Thirdly, it converts physical property units and unifies physical property data in different unit systems into SI units or other selected standard unit systems. For example, temperature is converted from Fahrenheit or Kelvin to Celsius, and pressure is converted from psi or bar to Pa.

[0029] The knowledge graph construction unit communicates with the data integration and standardization unit to construct a petrochemical material knowledge graph based on standardized data. This unit first extracts relationships between material entities and physical properties, identifying chemical substances, physical property parameters, and their relationships from text and structured data. Next, it constructs substance-property-condition triples, forming the basic units of the knowledge graph. For example, (n-heptane, density, 20°C / 1 atm) = 0.684 g / cm³. Finally, it implements knowledge reasoning and completion, using known relationships to infer unknown knowledge and fill in data gaps. For example, given the density at two temperatures, the density at the intermediate temperature can be inferred through linear interpolation.

[0030] In one embodiment of the present invention, the data processing and management module 1 employs a distributed data storage architecture, using MongoDB to store unstructured data, PostgreSQL to store structured data, and Neo4j to store knowledge graphs. Internally, the module uses a message queue (such as Kafka) for data transmission, ensuring efficient and reliable data processing. Furthermore, the module implements a version control mechanism that records all data changes and supports data backtracking and comparison.

[0031] The multi-scale feature characterization module 2 includes a quantum chemical feature characterization unit, a molecular structure feature characterization unit, a mesoscopic scale feature characterization unit, a macroscopic thermodynamic feature characterization unit and a multi-scale feature fusion unit.

[0032] The quantum chemical characterization unit is used to construct molecular features at the quantum scale. In one embodiment of the present invention, this unit uses density functional theory (DFT) calculations to extract molecular electronic structure features, such as frontier molecular orbital (HOMO and LUMO) energy levels, energy gaps, and electron density distribution. Simultaneously, it calculates multipole moment parameters, such as the molecular dipole moment and quadrupole moment, to characterize the molecular polarity and charge distribution. Furthermore, it characterizes intermolecular interactions and calculates different types of intermolecular interaction energies, such as electrostatic interaction energy, dispersion forces, and hydrogen bonding energies.

[0033] The quantum chemical characterization unit preferably uses the B3LYP / 6-31G(d) computational level to perform full quantum chemical calculations on small molecules and functional groups, while semi-empirical methods or molecular fragment methods are used for approximate calculations on large molecules. Taking the ethanol molecule as an example, the HOMO energy level calculated by this unit is -6.52eV, the LUMO energy level is -0.85eV, the energy gap is 5.67eV, the dipole moment is 1.69D, the number of hydrogen bond donors is 1, and the number of hydrogen bond acceptors is 1. These quantum characteristics reflect the properties of molecules at the electronic level and are important for predicting intermolecular interactions and specific physical properties (such as polarity-related properties).

[0034] The Molecular Structure Characterization Unit communicates with the Quantum Chemical Characterization Unit to construct features at the molecular scale. This unit first extracts molecular topological features, representing the molecule as an atom-bond connection graph and calculating topological indices such as the Wiener index and Balaban index. Next, it constructs molecular geometric features, characterizing the arrangement of atoms in three-dimensional space and calculating shape descriptors such as sphericity and ellipticity, as well as surface area and volume. Finally, it identifies and characterizes functional groups such as hydroxyl, carboxyl, and ester groups, counting the number of occurrences and locations of each type. Finally, it generates molecular fingerprints, such as the Extended Connectivity Fingerprint (ECFP) and the MACCS bond fingerprint, for use in molecular similarity comparisons.

[0035] In one embodiment of the present invention, for the n-hexane molecule, the features extracted by the molecular structure characterization unit include: 6 carbon atoms, 14 hydrogen atoms, 19 single bonds, 0 rings, 5 rotatable bonds, 0 aromaticity, 0 topological polar surface area, 0 Ų, and a Wiener index of 36. These structural features intuitively reflect the molecular skeleton structure and connection mode, and are important for predicting physical properties such as boiling point and melting point.

[0036] The mesoscale characterization unit communicates with the molecular structure characterization unit to construct features at the mesoscale. This unit first characterizes molecular aggregates, describing the packing and arrangement of molecules at the mesoscale and predicting self-assembly propensities and interfacial properties. Secondly, it extracts phase behavior features, predicts the likelihood of a mixture forming multiple phases, and characterizes critical parameters and phase diagrams. Finally, it constructs statistical thermodynamic features to describe the spatial distribution of molecules, velocity distribution, and correlation function characteristics.

[0037] The mesoscopic characterization unit preferably uses molecular dynamics and Monte Carlo simulations to generate molecular aggregate structures and then extracts statistical mechanical features such as radial distribution functions and structure factors. For example, for a n-hexane-benzene mixture, this unit can simulate the phase behavior of the two components at different temperatures and concentrations, calculate mixing enthalpy and excess volume, and predict possible phase separation conditions. These mesoscopic features are important for predicting phase equilibrium and transport properties.

[0038] The macroscopic thermodynamic characterization unit communicates with the mesoscopic characterization unit to construct features at the macroscale. This unit first constructs the equation of state parameter features, including critical parameters (temperature, pressure, volume), eccentricity factors, polarity factors, and other parameters used to describe the PVT relationship. Next, it extracts thermodynamic property features, such as phase transition parameters (heat of vaporization, heat of fusion, etc.), entropy capacity parameters, and free energy parameters. Finally, it characterizes transport properties, such as viscosity parameters, thermal conductivity characteristics, and diffusion coefficient characteristics. Finally, it constructs mixing rule parameters, such as binary interaction parameters and activity coefficient model parameters.

[0039] In one embodiment of the present invention, the macroscopic thermodynamic characterization unit constructs the equation of state parameters for n-hexane, including the following: critical temperature Tc = 507.6K, critical pressure Pc = 3.025 MPa, critical volume Vc = 368 cm³ / mol, and eccentricity factor ω = 0.301. These macroscopic characteristics are directly relevant for equation of state calculations and physical property predictions.

[0040] The multi-scale feature fusion unit communicates with each of the aforementioned feature representation units to effectively fuse features at different scales. This unit first performs feature correlation analysis to identify relationships between features at different scales, eliminating redundant information while retaining complementary information. Next, it performs feature selection and dimensionality reduction, employing filtering, wrapper, or embedded feature selection methods to screen key features and reducing feature dimensionality through methods such as principal component analysis (PCA) and t-SNE. Finally, it performs feature transformation and combination, standardizing, normalizing, and performing nonlinear transformations on features to construct high-order features. Finally, it implements adaptive feature weight allocation, dynamically adjusting the weights of features at different scales based on the prediction task and physical property type.

[0041] Preferably, the multi-scale feature fusion unit employs a feature fusion method based on an attention mechanism, assigning different weights to features at different scales. For example, for density prediction, macroscopic thermodynamic features might receive a weight of 0.5, molecular structure features a weight of 0.3, and quantum chemical features a weight of 0.2. For polarity-related properties (such as the dielectric constant), the weight of quantum chemical features might be significantly increased. This adaptive weighting mechanism enables the system to flexibly adjust the importance of features based on the physical nature of different properties, thereby improving prediction accuracy.

[0042] Through the feature representation and fusion of the above four levels, the multi-scale feature representation module 2 can comprehensively capture the structural and property characteristics of petrochemical materials from micro to macro, providing rich, multi-level feature representation for subsequent physical property prediction.

[0043] The physical constraint modeling module 3 includes a thermodynamic constraint embedding unit, a state equation constraint unit, a transport process constraint unit and a physical constraint optimization integration unit.

[0044] The Thermodynamic Constraint Embedder is used to embed the fundamental laws of thermodynamics into the model structure. This unit first implements the energy conservation constraint of the first law of thermodynamics, ensuring that the change in the internal energy of the system is equal to the algebraic sum of heat and work, which can be expressed as:

[0045] ,

[0046] in: is the change in internal energy, unit is J; is the heat absorbed by the system, in J; The unit is J. A positive value indicates that the system absorbs heat or does work, while a negative value indicates that the system releases heat or the outside world does work on the system. Secondly, this unit implements the entropy increase principle of the second law of thermodynamics to ensure that the entropy of an isolated system always increases, that is:

[0047] ,

[0048] in: is the total entropy change, expressed in J / K, and includes the sum of the system entropy change and the environmental entropy change. This constraint ensures that the direction of any spontaneous process conforms to the principle of entropy increase and is a fundamental constraint on the evolution of physical systems. Furthermore, this unit embeds the phase equilibrium condition constraint, which ensures that the temperature, pressure, and chemical potential of each phase in a multiphase system are equal:

[0049] ,

[0050] ,

[0051] ,

[0052] in: For the The temperature of the phase, in K; For the Phase pressure, in Pa; For the Phase components The chemical potential is expressed in J / mol. This set of constraints is necessary for a multiphase system to reach thermodynamic equilibrium. Finally, the unit verifies thermodynamic consistency relations, such as Maxwell's relation, to ensure that the derivative of the thermodynamic state function satisfies:

[0053] ,

[0054] ,

[0055] in: is the temperature in K; is the volume, in units of ; is pressure, unit is Pa; is entropy, unit is J / K; Indicates in variables Under the same conditions, right These relations are the direct result of the thermodynamic state function satisfying the fully differential property. Preferably, the thermodynamic constraint embedding unit adopts the Lagrange multiplier method to incorporate these constraints into the loss function, that is, the original loss function Modify to:

[0056] ,

[0057] in: is the original loss function, which represents the error between the predicted value and the true value; is the Lagrange multiplier, which is a non-negative scalar parameter that weighs the original loss and the importance of the constraint; is the constraint condition, which indicates the degree of violation of the physical constraint and should ideally be zero; Represents all constraints For example, for a phase equilibrium constraint, you can set , minimizing the square of the chemical potential difference. The state equation constraint unit communicates with the thermodynamic constraint embedding unit to embed the material state equation constraints. This unit first embeds the cubic state equation constraints, such as the Peng-Robinson equation:

[0058] ,

[0059] in: is pressure, unit is Pa; is the gas constant, which is 8.314 J / (mol-K); is the temperature in K; is the molar volume, in m3 / mol; and are parameters related to intermolecular forces and molecular size, with units of Pa-m and mol is the temperature correction function, dimensionless, usually expressed as:

[0060] ,

[0061] in: is the reduced temperature, dimensionless; is the critical temperature, in K; and the eccentricity factor The relevant parameters, dimensionless, can be obtained by calculate.

[0062] Secondly, the unit embeds the SAFT series of advanced state equation constraints, such as the PC-SAFT equation, which can more accurately describe chain molecules and hydrogen bond systems. Thirdly, it implements mixing rule constraints to define the state equation parameter calculation method for multi-component systems, such as the quadratic mixing rule:

[0063] ,

[0064] ,

[0065] in: and is the mixture state equation parameter, which has the same and Same units; and For components and The mole fraction of , dimensionless; 、 、 is the pure component parameter; is the binary interaction parameter, dimensionless, characterizing the components and The degree to which the interactions between them deviate from the geometric mean; and For all components and Perform the summation.

[0066] Finally, the unit performs special conditional equation of state corrections, such as critical region corrections, high pressure corrections, and very low temperature corrections, to improve prediction accuracy under special conditions. For example, critical region corrections can use renormalization group theory to correct critical exponents, avoiding inaccuracies in the classical equation of state near critical points.

[0067] The Transport Constraints unit communicates with the Thermodynamic Constraint Embedding unit to embed physical constraints describing material transport phenomena. This unit first embeds viscosity model constraints, such as theoretical constraints for low-density gas viscosity (based on the Chapman-Enskog theory) and empirical relations for liquid viscosity (such as the Andrade equation). Next, it embeds heat conduction constraints to ensure that heat conduction predictions conform to Fourier's law. Third, it embeds diffusion coefficient constraints to implement Fick's diffusion law and the Maxwell-Stefan diffusion model. Finally, it embeds surface property constraints, such as theoretical constraints for surface tension and the temperature dependence of surface tension (Eötvös's rule).

[0068] In one embodiment of the present invention, the transport process constraint unit uses the modified Andrade equation constraint on the liquid viscosity: ,

[0069] in: is the liquid viscosity, in Pa·s; is the temperature in K; 、 and is a substance-related parameter, where is a dimensionless constant, The unit is K, is a dimensionless constant. For n-hexane, the parameter value is approximately This equation reflects the exponential relationship between the viscosity of a liquid and its temperature and is an important constraint on its transport properties.

[0070] The physical constraint optimization and integration unit communicates with each of the aforementioned constraint units and is responsible for optimizing and integrating multiple physical constraints. This unit first optimizes the weights of each physical constraint, adjusting the importance of constraints based on the physical properties and prediction task. Secondly, it combines the hard and soft constraint mechanisms, assigning strictly enforced physical laws as hard constraints and empirical rules as soft constraints. Thirdly, it coordinates multi-scale constraint relationships to ensure consistency across physical constraints. Finally, it evaluates the effectiveness of constraint embedding, monitoring changes in constraint satisfaction and prediction performance.

[0071] The physical constraint optimization integrated unit preferably employs an adaptive constraint weight adjustment strategy, dynamically adjusting weights based on the degree of constraint violation. For example, when a model severely violates energy conservation, the corresponding constraint weight is automatically increased; when the constraints are already well satisfied, the weight can be appropriately reduced to prevent excessive constraints from impacting model flexibility. Furthermore, the unit employs a phased constraint implementation strategy, first training the model to satisfy basic physical constraints, then gradually adding more complex constraints to avoid optimization difficulties.

[0072] Through the collaborative work of the above units, the physical constraint modeling module 3 can ensure that the physical property prediction results conform to the basic physical laws and thermodynamic laws, avoid physically unreasonable prediction results, and improve the reliability and interpretability of the model.

[0073] The component interaction modeling module 4 includes a binary component interaction unit, a multi-component interaction unit, a component interaction representation network unit and a non-ideal mixed effect unit.

[0074] The binary component interaction unit is used to accurately capture the interaction between two components. The unit first calculates the binary interaction parameters and predicts the strength of the interaction between the components based on molecular similarity, interaction energy and solubility parameters. It can be estimated by the following equation:

[0075] ,

[0076] in: is the binary interaction parameter, dimensionless, characterizing the components and The degree to which the interactions between them deviate from the geometric mean; and is the attraction parameter of the pure component, in Pa·m , is related to the critical parameters. When , it means that the interaction between components follows the geometric mean rule; Indicates that the interaction between components is weaker than the geometric mean; Indicates that the interactions between the components are stronger than the geometric mean. Secondly, the unit predicts the excess properties of binary mixtures, such as excess enthalpy , excess volume and excess Gibbs free energy These excess properties reflect the degree to which the mixture deviates from ideal behavior. For example, the excess enthalpy can be expressed as:

[0077] ,

[0078] in: is the excess enthalpy, in J / mol; and are the mole fractions of the two components, dimensionless, and satisfy ; is the fitting parameter, the unit is J / mol; is the polynomial order, usually 2-4; Indicates from arrive This equation is a form of the Redlich-Kister expansion and is often used to characterize the excess properties of binary mixtures.

[0079] Thirdly, the unit predicts binary phase diagrams, including liquid-liquid equilibrium, gas-liquid equilibrium, and azeotropic point predictions, etc. Finally, it simulates special binary interaction effects such as hydrogen bonding, charge transfer, acid-base interaction, and complexation.

[0080] In one embodiment of the present invention, for the ethanol-water system, the excess enthalpy predicted by the binary component interaction unit is negative at 25°C, indicating that the mixing process is exothermic, which is consistent with the fact that hydrogen bonds are formed between the two components; while for the hexane-benzene system, the predicted excess enthalpy is positive, indicating that the mixing process is endothermic, which is due to the weakened intermolecular interaction caused by the difference in the two molecular structures.

[0081] The multi-component interaction unit communicates with the binary component interaction unit to handle complex interactions between three or more components. This unit first identifies higher-order interaction terms, determines whether there are interactions unique to three or more components, and how to assess the importance of these higher-order terms. Higher-order interaction terms can be represented as follows:

[0082] ,

[0083] in: is the excess Gibbs free energy, in J / mol; 、 and For components 、 and The mole fraction of , dimensionless; is the binary interaction energy parameter, in J / mol, characterizing the components and the interactions between them; is the ternary interaction energy parameter, in J / mol, characterizing the components 、 and synergy between 、 and The first term on the right side of the equation represents all possible two-way interactions, the second term represents all possible three-way interactions, and so on.

[0084] Secondly, the unit optimizes multicomponent mixing rules, generalizing from binary parameters to multicomponent systems and accounting for the influence of concentration on the mixing rules. Thirdly, it predicts multicomponent phase behavior, identifying multiphase regions, critical endpoints, and azeotropic multicomponents. Finally, it simulates multicomponent dynamic interactions, predicting the effects of changes in component concentration, temperature, and pressure on these interactions.

[0085] The multicomponent interaction unit preferably uses a hierarchical approach to multicomponent interactions: first, all possible binary interactions are considered, and then the need for higher-order interactions is evaluated. For most systems, binary interactions provide a sufficiently accurate description; however, for some special systems (such as mixtures of highly polar components), the inclusion of ternary interactions can significantly improve prediction accuracy.

[0086] The component interaction representation network unit communicates with the multi-component interaction unit and is used to capture complex interactions between components through a neural network. This unit first constructs a component graph representation learning framework, representing the mixture as a network of component nodes. It extracts local interactions between components through graph convolution and focuses on important component interactions through a graph attention mechanism. Secondly, it implements a concentration-aware interaction network, adjusting interaction strength based on component concentration to handle extreme high and low concentration conditions. Thirdly, it constructs a condition-sensitive interaction model to capture the effects of conditions such as temperature and pressure on component interactions. Finally, it captures long-range interactions, overcoming local interaction limitations through attention mechanisms and global information aggregation.

[0087] In one embodiment of the present invention, the component interaction representation network unit uses a graph attention network (GAT) to construct a component interaction model, where the attention coefficient The calculation is as follows:

[0088] ,

[0089] in: is the attention coefficient, dimensionless, indicating the importance of component j on component i; and is the component node feature vector, with dimension , d is the feature dimension; W is the weight matrix, the dimension is , is the feature dimension after conversion; a is the attention vector, the dimension is ; Represents vector concatenation operation; is the neighbor set of component i; exp represents the exponential function; LeakyReLU represents the leaky rectified linear unit activation function, defined as LeakyReLU(x)=max(0.01x,x); represents the sum of all k neighbors of component i. In this way, the model can automatically learn the contribution of different components to the interaction. The non-ideal mixture effect unit A4 is connected to the component interaction characterization network unit A3 to accurately describe the mixture effect that deviates from the ideal behavior. This unit first implements the activity coefficient model, such as NRTL, UNIQUAC and Wilson equation, to predict the non-ideal behavior of components in the mixture. For example, the expression for calculating the activity coefficient of the NRTL equation is:

[0090] ,

[0091] in: For components The activity coefficient of , dimensionless; For components The mole fraction of , dimensionless; is the energy parameter, dimensionless, usually expressed as ,in For components and The interaction energy parameter between them is in J / mol; is the NRTL model parameter, dimensionless, usually expressed as ,in It is a non-random parameter, dimensionless, usually between 0.1 and 0.5; 、 and represents the sum over all components. This equation is based on local composition theory, which takes into account the difference between the local environment surrounding a molecule and the global environment.

[0092] Secondly, the unit comprehensively models excess properties, capturing the correlations between different excess properties and predicting the effects of temperature and pressure on them. Thirdly, it handles special non-ideal systems such as strong electrolyte systems, polymer solutions, and supercritical fluid mixtures. Finally, it predicts phase separation phenomena, implements thermodynamic stability criteria, and calculates the boundaries of the two-phase region and the critical solution temperature.

[0093] The non-ideal mixing effect unit automatically selects the most appropriate activity coefficient model based on the mixture's characteristics. For example, for fully miscible systems, the Wilson equation is generally effective; for partially miscible systems, the NRTL and UNIQUAC models are more suitable; and for polymer solutions, the Flory-Huggins model may be appropriate. Furthermore, the unit implements a hybrid model approach, combining the strengths of different activity models to improve the applicability and accuracy of predictions.

[0094] Through the collaborative work of the above units, the component interaction modeling module 4 can accurately capture the complex interactions in petrochemical material mixtures, predict non-ideal mixing effects, and provide key support for physical property prediction.

[0095] The uncertainty quantification module 5 includes a prediction uncertainty source analysis unit, a Bayesian uncertainty quantification unit, an uncertainty propagation and aggregation unit, and an uncertainty calibration and evaluation unit.

[0096] The prediction uncertainty source analysis unit is used to identify and analyze the sources of uncertainty. This unit first evaluates data uncertainty, including measurement errors in experimental data, uncertainty caused by data sparsity, and data noise. The measurement uncertainty of experimental data can be obtained from the documentation of the data source or estimated based on the standard deviation of repeated measurements. Secondly, the model uncertainty is analyzed, including the uncertainty caused by the choice of model structure, the uncertainty of parameter estimation, and the risk of overfitting or underfitting caused by model complexity. Thirdly, the physical approximation uncertainty is quantified, and the uncertainty introduced by factors such as state equation approximation, mixing rule simplification, and quantum computing accuracy limitations is evaluated. Finally, the uncertainty of extrapolation of prediction conditions is evaluated, and the reliability is analyzed when the prediction conditions (such as temperature, pressure, and composition) exceed the range of training data.

[0097] In one embodiment of the present invention, the prediction uncertainty source analysis unit predicts liquid density and finds that the data uncertainty at normal temperature and pressure is about ±0.5%, which mainly comes from experimental measurement errors; under high temperature and high pressure conditions, the uncertainty may increase to more than ±2%, which is mainly due to data sparsity and model extrapolation.

[0098] The Bayesian uncertainty quantification unit is connected to the prediction uncertainty source analysis unit and is used to implement uncertainty quantification using the Bayesian method. This unit first implements a Bayesian neural network, replaces the deterministic weights with probability distributions, and represents parameter uncertainty through the posterior distribution. In the Bayesian neural network, the posterior distribution of the prediction y can be expressed as:

[0099] ,

[0100] in: For a given input and training data Next, output The posterior probability distribution of ; is the input feature vector; is the training data set; is the model parameter vector; is the parameter posterior distribution, which represents the probability distribution of the parameter under given training data; is the predicted distribution under given parameters, usually a Gaussian distribution; represents the integration over all possible parameter spaces. This integral cannot usually be calculated analytically and needs to be approximated by numerical methods such as variational inference or Monte Carlo sampling. Secondly, this unit implements the Monte Carlo Dropout technique, which obtains the prediction distribution by keeping Dropout turned on during the prediction phase and making multiple random predictions. The prediction mean and variance can be estimated as follows:

[0101] ,

[0102] ,

[0103] in: is the predicted mean; Var is the prediction variance; For the The model prediction of the dropout sampling is the forward propagation result of the network after randomly blocking some neurons; is the intrinsic variance of the model prediction, i.e., the variance of the assumed observation noise; is the number of sampling times, usually set to 50-100; Express The sub-sampling results are averaged.

[0104] Thirdly, the unit builds Bayesian ensemble learning, combining the predictions of multiple Bayesian models to further improve the reliability of uncertainty estimation. Finally, it implements deep Gaussian processes, constructing kernel functions through deep networks to handle Gaussian process regression on large datasets.

[0105] Preferably, the Bayesian uncertainty quantification unit uses different uncertainty quantification strategies for different types of properties. For example, for data-rich properties (such as density at normal temperature and pressure), a simple Monte Carlo dropout method can be used; while for data-sparse or highly nonlinear properties (such as properties near critical points), a full Bayesian neural network or deep Gaussian process method may be required.

[0106] The Uncertainty Propagation and Aggregation Unit communicates with the Bayesian Uncertainty Quantification Unit to track the propagation of uncertainty during the prediction process. This unit first analyzes the uncertainty propagation chain, tracking how input variable uncertainty propagates to the prediction results, assessing intermediate feature uncertainties, and analyzing the accumulation of uncertainty in multi-step predictions. Second, it aggregates multi-source uncertainty, comprehensively considering various sources of uncertainty such as data, models, and physical approximations, and identifying the main uncertainty contributions. Third, it analyzes condition-dependent uncertainty, studying the impact of conditions such as temperature, pressure, and composition on prediction uncertainty. Finally, it addresses extreme conditional uncertainty, addressing high uncertainty in special cases such as near critical points, phase transition regions, and trace components.

[0107] In one embodiment of the present invention, the uncertainty propagation and aggregation unit discovered that when predicting the viscosity of hydrocarbon mixtures, uncertainty increases significantly as the temperature approaches the critical point, with the relative error potentially increasing from ±5% under normal conditions to over ±20%. This is primarily due to the highly nonlinear and sensitive physical behavior in the critical region, which challenges the model's generalization capabilities.

[0108] The Uncertainty Calibration and Evaluation Unit communicates with the Uncertainty Propagation and Aggregation Unit to ensure the accuracy of uncertainty estimates. This unit first implements uncertainty calibration methods, such as frequency calibration and temperature scaling, to ensure that the coverage of the prediction interval is consistent with the confidence level. The goal of frequency calibration is to ensure that:

[0109] ,

[0110] in: represents probability; is the true value; is the predicted mean; is the prediction standard deviation; is the calibration factor, which is a standard normal distribution Quantile; The goal of calibration is to ensure that the actual coverage (i.e., the proportion of true values ​​falling within the prediction interval) is equal to the nominal coverage (i.e., 1- .

[0111] Second, the unit calculates reliability assessment metrics, such as prediction interval coverage, negative log-likelihood, and calibration error, to comprehensively evaluate the quality of uncertainty estimates. Third, it implements uncertainty visualization, visually displaying the uncertainty distribution through prediction distribution plots, uncertainty heatmaps, and confidence interval strip charts. Finally, it detects anomalous predictions, flags high-uncertainty predictions, identifies unusual distribution shapes, and analyzes sudden changes in uncertainty.

[0112] The uncertainty calibration and evaluation unit preferably employs a tiered calibration strategy, calibrating different condition intervals separately to improve calibration accuracy. For example, the temperature range can be divided into three intervals: low, medium, and high, with the uncertainty estimate calibrated separately for each interval. Furthermore, the unit implements an adaptive recalibration mechanism, regularly updating calibration parameters as new data accumulates, ensuring that uncertainty estimates remain accurate.

[0113] Through the collaborative work of the above units, the uncertainty quantification module 5 can provide reliable forecast uncertainty assessment, quantify the credibility of the forecast results, and support risk-based decision making.

[0114] The physical property prediction and analysis module 6 includes a thermodynamic property prediction unit, a transport property prediction unit, a combustion property prediction unit and a structure-property correlation analysis unit.

[0115] The Thermodynamic Property Prediction Unit is used to predict various thermodynamic properties. This unit first predicts state properties, including basic thermodynamic properties such as density, specific heat capacity, enthalpy, and entropy. For example, liquid density prediction can be achieved using the modified Rackett equation:

[0116] ,

[0117] in: is the liquid density in kg / m 3 is the critical density, in kg / m 3 is the critical compressibility factor, dimensionless, defined as ,in 、 and are the critical pressure, critical molar volume, and critical temperature, respectively; is the reduced temperature, dimensionless, defined as ,in is the system temperature; the superscript 2 / 7 denotes the power exponent. This equation, based on the critical point scaling principle, accurately predicts the density of liquids from low temperatures to near the critical point. Secondly, this unit predicts phase equilibrium properties, including vapor pressure, phase equilibrium composition, and critical parameters. For example, the vapor pressure of a pure component can be predicted using the Antoine equation:

[0118] ,

[0119] in: is the saturated vapor pressure, in Pa; is the temperature in K; 、 and is the material-related constant, where is a dimensionless constant, The unit is K, The unit is K. This equation is an empirical relationship that accurately predicts the vapor pressure of pure components over a limited temperature range.

[0120] Next, the unit calculates thermodynamic derivative properties such as expansion coefficient, compressibility, and coke-tan coefficient. Finally, it predicts mixing properties such as mixing enthalpy, excess volume, and activity coefficient.

[0121] In one embodiment of the present invention, the thermodynamic property prediction unit predicts the liquid density of C1-C10 normal alkanes in the temperature range of 283.15-373.15K and the pressure range of 0.1-10MPa with an average relative error of less than 0.5%, surpassing traditional equation of state methods (typically 1%-3%). For complex mixtures, such as crude oil, the unit can predict density based on component composition with an error of less than 1%.

[0122] The transport property prediction unit communicates with the thermodynamic property prediction unit and is used to predict various transport properties. This unit first predicts viscosity, including gas viscosity, liquid viscosity, and high-pressure viscosity. For example, liquid viscosity prediction can be achieved using the free volume theory model:

[0123] ,

[0124] in: is the liquid viscosity, in Pa·s; The reference viscosity is in Pa·s, usually the viscosity value at the extreme low density is taken. is the substance-related constant, with the unit being m3 / mol; is the free volume, in m3 / mol, which is related to temperature and density and can be expressed as ,in is the actual molar volume, is the molar volume when densely packed. This model is based on the free volume theory of molecular motion and can describe the changes in liquid viscosity with temperature and pressure.

[0125] Secondly, the unit predicts thermal conductivity and estimates the thermal conductivity of gases and liquids. Thirdly, the diffusion coefficient is calculated to predict the self-diffusion coefficient and the mutual diffusion coefficient. Finally, surface properties such as surface tension and interfacial tension are predicted. Preferably, the transport property prediction unit uses a multi-scale fusion method to predict viscosity, combining molecular dynamics simulation and empirical correlation to improve prediction accuracy. For hydrocarbon mixtures, the average relative error of the viscosity prediction of this unit in the temperature range of 293.15~473.15K is 3%~5%, which is significantly better than the traditional corresponding state method (usually 10%~15%). The combustion property prediction unit is communicated with the thermodynamic property prediction unit to predict properties related to combustion. The unit first predicts the fuel evaluation index, including octane number, cetane number and methane number. For example, the research octane number (RON) of gasoline can be predicted by the component composition:

[0126] ,

[0127] Where: RON is the research octane number of the mixture, dimensionless; and For components and Volume fraction of, dimensionless; Pure components The octane number, dimensionless; For components and The binary interaction term of is dimensionless and represents the nonlinear effect of the two-component mixing on the octane number; and This equation takes into account the linear contribution and binary interaction effects of the components and can accurately predict the octane number of the mixed fuel.

[0128] Secondly, the unit predicts fire safety properties such as flash point, ignition point, and explosion limits. Thirdly, it calculates combustion thermodynamic properties such as heat of combustion and adiabatic flame temperature. Finally, it predicts combustion kinetic parameters such as ignition delay time and laminar flame speed.

[0129] In one embodiment of the present invention, the combustion property prediction unit predicts the research octane number (RON) of gasoline components with an average absolute error of 1.5 to 2.0 units, and predicts the cetane number of diesel with an average absolute error of 1.0 to 1.5 units. This accuracy meets the requirements of most industrial applications.

[0130] The structure-property correlation analysis unit communicates with each of the above prediction units and analyzes the relationship between molecular structure and physical properties. This unit first identifies key structural features, identifying key structural fragments and functional groups that influence physical properties. Next, it predicts the effects of structural modifications, analyzing substituent effects, homologous variation trends, and positional isomerism. Third, it analyzes component interaction mechanisms, identifies synergistic and antagonistic effects, and constructs a network of component interactions. Finally, it analyzes the drivers of property changes, quantifies the impact of each factor on the properties, and plots the response curves of the properties to the conditions.

[0131] The structure-property association analysis unit preferably employs interpretable artificial intelligence techniques, such as SHAP value analysis, to quantify the contribution of different molecular features to property predictions. For example, for boiling point prediction of alkanes, the unit found that the number of carbon atoms was the most important factor (contributing approximately 75%), the degree of branching was the second most important factor (contributing approximately 15%), and the presence of ring structures was the third most important factor (contributing approximately 5%). This analysis not only improves the interpretability of predictions but also provides valuable guidance for molecular design.

[0132] Through the collaborative work of the above units, the physical property prediction and analysis module 6 can accurately predict the various physical and chemical properties of petrochemical materials, analyze the relationship between molecular structure and physical properties, and provide a scientific basis for process optimization and product development.

[0133] In a preferred embodiment of the present invention, the system also includes a model training and optimization module 7, which includes a multi-task joint learning unit, a physical constraint optimization unit, a model architecture optimization unit, a hyperparameter optimization unit and a model evaluation and selection unit.

[0134] The multi-task joint learning unit is used to achieve joint prediction of multiple physical properties. This unit first designs the physical property association structure, establishes the dependency relationship between physical properties, and designs the shared feature layer and the physical property specific layer. For example, physical properties such as density, viscosity, and surface tension can share low-level feature representations because they are all closely related to intermolecular forces. Secondly, a multi-task loss function is designed, which assigns weights based on physical property importance and prediction difficulty, and takes into account data uncertainty and consistency between physical properties. The multi-task loss function can be expressed as:

[0135] ,

[0136] in: is the total loss function of multi-task; For the task The loss function is usually the mean square error or mean absolute error; For the task The weight, dimensionless, reflects the importance of the task; For the task and The consistency constraint between them represents the physical relationship that the prediction results of the two tasks should satisfy; is the constraint weight, dimensionless; For all tasks Perform summation; Indicates that all tasks Perform the summation.

[0137] Next, the unit implements a gradient balancing strategy to normalize the gradients of different tasks, detect and resolve gradient conflicts, and dynamically adjust task priorities. Finally, it implements a transfer and incremental learning strategy, leveraging previously learned physical properties to assist in learning new physical properties and avoid catastrophic forgetting.

[0138] The multi-task joint learning unit preferably employs a soft parameter sharing approach, where the models for different tasks have their own parameters, but regularization constraints are imposed between these parameters to facilitate knowledge sharing between tasks. For closely related properties (such as density and viscosity), the degree of parameter sharing is high; for less correlated properties (such as octane number and surface tension), the degree of parameter sharing is low. This flexible sharing mechanism enables the model to fully exploit the inter-task correlation while maintaining sufficient flexibility.

[0139] The physical constraint optimization unit optimizes the implementation of physical constraints in the model. This unit first designs a constraint-weighted loss function to balance data-driven loss with physical constraint loss, adjusting the penalty strength based on the degree of violation. Next, it implements the Lagrange multiplier method to handle equality and inequality constraints and designs an efficient multiplier update strategy. Third, it designs a physics-guided regularization method that leverages prior knowledge to constrain the model, ensuring physical invariance and gradually increasing constraint strength. Finally, it constructs a constraint satisfaction feedback mechanism to quantify constraint satisfaction, dynamically adjust constraint priorities, and identify areas where constraints are difficult to satisfy.

[0140] In one embodiment of the present invention, the physical constraint optimization unit adopts a soft constraint method for the energy conservation constraint of the first law of thermodynamics, and adds the constraint violation degree to the loss function:

[0141] ,

[0142] in: is the modified total loss function; It is the data-driven loss, usually the mean squared error between the predicted value and the true value; is the constraint weight, dimensionless, controlling the importance of the constraint; MSE is the mean square error function, defined as MSE ,in is the sample size; 、 and are the change in internal energy, heat, and work respectively. Constraint weights The initial setting is 0.1 and gradually increases to 1.0 as training progresses to avoid overly strong constraints in the early stages of training that may lead to optimization difficulties.

[0143] The Model Architecture Optimization unit is responsible for optimizing the architectural design of neural networks. This unit first optimizes graph neural networks, improving the graph convolutional layers, graph attention mechanisms, and graph pooling strategies, and optimizing the representation of component interactions. Secondly, it designs a multi-scale feature fusion network, constructs a cross-scale feature aggregation mechanism, optimizes scale-specific subnetworks, and implements adaptive feature weighting. Thirdly, it constructs a physical embedding neural network architecture, designing the physical equation layer, state variable transformation layer, and physical consistency check layer. Finally, it implements an adaptive architecture search to automatically discover the optimal network structure, balancing model complexity and performance, and optimizing the architecture for different prediction tasks.

[0144] The model architecture optimization unit preferably uses Neural Architecture Search (NAS) technology to automatically search for the optimal network structure within a predefined search space. This search space includes hyperparameters such as the number of layers, layer width, activation function, connection method, and attention mechanism. By using efficiency-optimized search algorithms (such as gradient-based methods or evolutionary algorithms), this unit can find a network architecture that balances performance and efficiency within reasonable computing resources.

[0145] The hyperparameter optimization unit optimizes the model's hyperparameter settings. This unit first performs an automated hyperparameter search, finding the optimal hyperparameters through grid search, random search, Bayesian optimization, or evolutionary algorithms. Next, it optimizes the learning rate strategy, designs a learning rate scheduling mechanism, implements an adaptive learning rate method, and adjusts the learning rate based on constraint violations. Third, it optimizes regularization parameters, adjusting the L1 / L2 regularization strength, dropout rate, and early stopping policy parameters. Finally, it optimizes model ensemble parameters, determining the number of ensemble models, controlling model diversity, optimizing ensemble weights, and selecting the optimal ensemble method.

[0146] In one embodiment of the present invention, the hyperparameter optimization unit uses the Bayesian optimization method to search for the optimal hyperparameters, and defines the search space as: the learning rate is in the range [1e -5 ,1e -2 ] range, the batch size is in the range of [16,256], and the L2 regularization coefficient is in the range of [1e -6 ,1e -3 ] range, and the Dropout rate is in the range of [0.1, 0.5]. By optimizing the validation set performance, the unit found the optimal hyperparameter combination: the learning rate is 3e -4 , the batch size is 64, and the L2 regularization coefficient is 5e -5 , the Dropout rate is 0.2.

[0147] The Model Evaluation and Selection Unit evaluates model performance and selects the optimal model. This unit first conducts a comprehensive multi-metric evaluation, calculating prediction accuracy metrics (such as MAE, RMSE, and R²), physical consistency assessment metrics, and uncertainty quality metrics. Next, it designs a cross-validation strategy, implementing multi-fold cross-validation, holdout validation, time series cross-validation, and stratified cross-validation. Finally, it conducts model comparison and selection, designing a model ranking mechanism, searching for Pareto optimal solutions, performing statistical significance tests, and introducing a model complexity penalty. Finally, it evaluates extrapolation capability, testing conditional extrapolation, component extrapolation, extreme condition prediction, and long-term prediction stability.

[0148] The model evaluation and selection unit preferably employs a multi-objective evaluation framework, simultaneously considering three objectives: prediction accuracy, physical consistency, and computational efficiency. For prediction accuracy, the root mean square error (RMSE) and coefficient of determination (R²) are used; for physical consistency, the degree of thermodynamic constraint violation is used; and for computational efficiency, inference time and the number of model parameters are used. Through Pareto front analysis, the unit identifies balanced models that perform well on all three objectives, rather than models that excel on only one metric.

[0149] Through the collaborative work of the above units, the model training and optimization module 7 can realize multi-task joint learning, optimize the implementation of physical constraints, optimize the model architecture and hyperparameters, evaluate model performance and select the optimal model, thereby improving the prediction accuracy and reliability of the system.

[0150] In a preferred embodiment of the present invention, the system further comprises a system integration and application module 8, which comprises a user interface unit, an industrial integration unit, a data exchange and standardization unit, a deployment and expansion unit and an application scenario adaptation unit.

[0151] The User Interface Unit is responsible for providing a user-friendly interactive interface. This unit first designs a web application interface, employing a responsive layout, constructing a physical property prediction form, enabling result visualization, and managing user permissions. Next, it provides a command-line interface that supports batch commands and scripted operations, designing advanced query syntax, and customizing output formats. Finally, it develops a programming API, designing a RESTful API, providing a Python client library and SDK, and implementing a real-time Web Socket interface. Finally, it builds a mobile application interface to optimize the mobile device experience, support offline predictions, and implement data synchronization and push notifications.

[0152] Preferably, the user interface unit adopts a front-end and back-end separation architecture, with the front-end using the React framework to build a responsive web interface, and the back-end providing RESTful API services. The user interface provides three main functional modes: (1) single-point prediction mode, in which the user enters a specific substance and condition, and the system returns the prediction result and uncertainty; (2) batch prediction mode, in which the user uploads a file containing multiple substances and conditions, and the system performs batch prediction and returns the results; (3) parameter optimization mode, in which the user sets the target property range, and the system recommends the optimal group ratio that meets the conditions.

[0153] The Industrial Integration Unit is used to integrate with industrial systems. This unit first integrates process simulation software, developing interfaces with Aspen Plus, ProII / Petro-SIM, and HYSYS, and constructing a gPROMS dynamic model interface. Secondly, it provides industrial control system interfaces, implements OPC UA servers and clients, integrates DCS and SCADA systems, and connects to MES systems. Thirdly, it integrates with laboratory information systems and designs LIMS interfaces to support automatic import of experimental data, generate analysis reports, and track sample information. Finally, it connects to enterprise resource planning systems, implements SAP / Oracle integration, connects to product data management and supply chain management systems, and integrates with quality management systems.

[0154] In one embodiment of the present invention, the Industrial Integration Unit has developed a plug-in for the Aspen Plus physical property package, allowing users to directly access the system's physical property prediction capabilities within Aspen Plus. This plug-in, implemented through Aspen's physical property extension interface, supports the calculation of over 15 physical property parameters, including density, vapor pressure, enthalpy, entropy, heat capacity, and viscosity. Integration testing has shown that the use of this plug-in significantly improves the convergence of Aspen Plus when processing complex mixtures, particularly for highly nonideal mixtures and supercritical conditions.

[0155] The Data Exchange and Standardization Unit is used to implement standardized data exchange. First, this unit supports multiple data exchange formats, including JSON / XML, CSV / Excel, HDF5 for large data, and CDF for scientific data. Second, it implements industry-standard interfaces, supporting the CAPE-OPEN standard, the OPC UA information model, the ASTM data standard, and the ISA-95 hierarchy. Third, it provides data conversion and mapping capabilities, enabling unit conversion, data schema mapping, encoding conversion, and coordinate system conversion. Finally, it ensures secure data transmission through data encryption, security authentication, access logging, and data integrity verification.

[0156] The Data Exchange and Standardization Unit preferably implements a thermodynamic interface based on the CAPE-OPEN standard, enabling seamless system integration with any process simulation software supporting this standard. The CAPE-OPEN interface provides standardized thermodynamic calculation services, including physical property calculations, phase equilibrium calculations, and thermodynamic derivative calculations. Furthermore, the unit supports common physical property data exchange formats, such as DIPPRXML and ThermoML, facilitating data exchange with other physical property databases and systems.

[0157] The Deployment and Expansion unit handles system deployment and expansion. First, this unit supports multi-environment deployment, providing cloud deployment solutions, edge computing deployment solutions, local server deployment solutions, and hybrid deployment strategies. Second, it implements containerization and microservices architecture, encapsulating Docker containers, orchestrating them using Kubernetes, designing microservice decomposition, and building a service mesh. Third, it optimizes performance and expansion, implements load balancing, supports horizontal scaling, optimizes caching, and implements distributed computing. Finally, it manages version control and upgrades, tracks system versions, enables smooth upgrades, designs rollback plans, and manages configurations.

[0158] In one embodiment of the present invention, the deployment and expansion unit utilizes a microservices architecture, breaking the system down into multiple independent services, including data processing, feature extraction, model inference, and result analysis. Each service is encapsulated in a Docker container and orchestrated and managed via Kubernetes. This architecture ensures high scalability and resilience, enabling automatic scaling of service instances based on load and rapid recovery from service failures. Tests have shown that even with high concurrency (100 requests per second), system response times remain below 200 milliseconds, meeting the requirements of real-time applications.

[0159] The application scenario adaptation unit is used to optimize for different application scenarios. First, this unit adapts to petroleum refining applications, developing crude oil evaluation tools, predicting fraction properties, optimizing mixed crude oil ratios, and predicting product quality. Secondly, it adapts to chemical process applications, optimizing separation processes, simulating reaction processes, designing heat exchange networks, and optimizing process conditions. Thirdly, it adapts to new material development, providing solvent screening tools, predicting polymer properties, designing composite materials, and optimizing material performance. Finally, it supports environmental and safety assessments, assessing environmental impacts, analyzing safety risks, predicting emission characteristics, and supporting emergency response.

[0160] Optimized for petroleum refining applications, the application scenario adaptation unit has developed a crude oil evaluation and formulation optimization function. This function allows users to input the basic properties of various crude oils (such as API gravity, sulfur content, etc.) and available quantities. The system then automatically recommends the optimal blending ratio, ensuring that the properties of the blended crude oil meet refinery requirements while minimizing costs. Tests have shown that using this function can reduce crude oil procurement costs by 2% to 3% while ensuring that product quality meets specification requirements. Furthermore, the unit also provides a formulation optimization function for petrochemical product development scenarios, which can recommend the optimal component ratio based on target physical properties (such as octane number and viscosity), accelerating the product development process.

[0161] Through the collaborative work of the above units, the system integration and application module 8 can provide a user-friendly interactive interface, realize integration with industrial systems, perform data exchange and standardization, support system deployment and expansion, and adapt to different application scenarios to improve the practicality and ease of use of the system.

[0162] The present invention also provides a physical information-constrained petrochemical material mixed modeling and physical property calculation method, comprising the following steps:

[0163] Step S1: collecting multi-source data of petrochemical materials, and performing cleaning, quality control and standardization on the multi-source data to obtain standard format data;

[0164] Specifically, this step first involves collecting petrochemical material data from a variety of sources, including experimental databases, theoretical calculation results, and industrial process data. This raw data is then cleaned and quality-controlled, including detecting and addressing outliers, verifying the data's thermodynamic consistency, and assigning a quality score. The cleaned data is then integrated into a standardized format, including unifying physical property data formats, standardizing molecular representations, and converting physical property units. Finally, a petrochemical material knowledge graph is constructed based on this standardized data, including extracting relationships between material entities and physical properties, constructing material-property-condition triples, and implementing knowledge reasoning and completion.

[0165] Step S2: Based on the standard format data, construct a multi-level petrochemical material feature representation from quantum scale to macro scale, and achieve the fusion of features at different scales;

[0166] Specifically, this step first constructs molecular features from the quantum scale, including extracting molecular electronic structure features, calculating molecular dipole moments and multipole moments, and characterizing intermolecular interactions. Then, features are constructed from the molecular scale, including extracting molecular topological structure features, constructing molecular geometric structure features, identifying and characterizing functional groups, and generating molecular fingerprints. Next, features are constructed from the mesoscopic scale, including characterizing molecular aggregate features, extracting phase behavior features, and constructing statistical thermodynamic features. Again, features are constructed from the macroscopic scale, including constructing state equation parameter features, extracting thermodynamic property features, characterizing transport property features, and constructing mixing rule parameters. Finally, effective fusion of features at different scales is achieved, including feature correlation analysis, feature selection and dimensionality reduction, feature conversion and combination, and feature adaptive weight allocation.

[0167] Step S3: Embed the thermodynamic laws and physical law constraints into the model structure to ensure that the physical property prediction results meet the thermodynamic consistency requirements;

[0168] Specifically, this step first embeds the basic laws of thermodynamics into the model structure, including implementing the energy conservation constraint of the first law of thermodynamics, implementing the entropy increase principle constraint of the second law of thermodynamics, embedding the phase equilibrium condition constraint and verifying the thermodynamic consistency relationship. Then, the material state equation constraints are embedded, including embedding the cubic state equation constraints, embedding the SAFT series advanced state equation constraints, implementing the mixing rule constraints and performing special condition state equation corrections. Next, the physical constraints describing the material transfer phenomenon are embedded, including embedding the viscosity model constraints, embedding the heat conduction constraints, embedding the diffusion coefficient constraints and embedding the surface property constraints. Finally, multiple physical constraints are optimized and integrated, including optimizing the weights of each physical constraint, combining soft and hard constraint mechanisms, coordinating multi-scale constraint relationships and evaluating the constraint embedding effect.

[0169] Step S4: Accurately capture the complex interactions between different components in the petrochemical material mixture, predict the non-ideal mixing effects between components, and build a component interaction knowledge base;

[0170] Specifically, this step first accurately captures the interaction between two components, including calculating binary interaction parameters, predicting binary mixing excess properties, predicting binary phase diagrams, and simulating special binary interaction effects. Then, it handles complex interactions between three or more components, including identifying high-order interaction terms, optimizing multi-component mixing rules, predicting multi-component phase behavior, and simulating multi-component dynamic interactions. Next, it captures complex interactions between components through neural networks, including constructing a component graph representation learning framework, realizing concentration-aware interaction networks, constructing condition-sensitive interaction models, and capturing long-range interactions. Finally, it accurately describes mixing effects that deviate from ideal behavior, including realizing activity coefficient models, comprehensively modeling excess properties, handling special non-ideal systems, and predicting phase separation phenomena.

[0171] Step S5: Evaluate the reliability of the physical property prediction results, quantify the sources and magnitude of the uncertainty in the prediction, and generate a confidence interval for the prediction results;

[0172] Specifically, this step first identifies and analyzes the sources of uncertainty, including assessing data uncertainty, analyzing model uncertainty, quantifying physical approximation uncertainty, and evaluating uncertainty in extrapolating prediction conditions. Uncertainty quantification is then achieved through Bayesian methods, including implementing Bayesian neural networks, implementing Monte Carlo dropout, constructing Bayesian ensemble learning, and implementing deep Gaussian processes. Next, the propagation of uncertainty throughout the prediction process is tracked, including analyzing uncertainty propagation chains, aggregating multiple sources of uncertainty, analyzing condition-dependent uncertainty, and handling uncertainty under extreme conditions. Finally, the accuracy of uncertainty estimates is ensured, including implementing uncertainty calibration methods, calculating reliability assessment metrics, implementing uncertainty visualization, and detecting anomalous predictions.

[0173] Step S6: Based on the above steps, various physical and chemical properties of petrochemical materials are predicted, the relationship between molecular structure and physical properties is analyzed, and visualization and interpretation functions are provided.

[0174] Specifically, this step first predicts various thermodynamic properties, including state properties, phase equilibrium properties, thermodynamic derivative properties, and mixing properties. Then, various transport properties are predicted, including viscosity, thermal conductivity, diffusion coefficient, and surface properties. Next, combustion-related properties are predicted, including fuel evaluation index, fire safety properties, combustion thermodynamic properties, and combustion kinetic parameters. Finally, the relationship between molecular structure and physical properties is analyzed, including identifying key structural features, predicting structural modification effects, analyzing component interaction mechanisms, and analyzing the driving factors of physical property changes. At the same time, intuitive result display and interpretation are provided, including multidimensional physical property visualization, contribution analysis visualization, uncertainty visualization, and interactive exploration tools.

[0175] The coordinated implementation of the above steps constitutes a complete physical information-constrained petrochemical material mixture modeling and physical property calculation method, which can accurately predict the various physical and chemical properties of petrochemical materials while providing reliable uncertainty assessment and intuitive interpretation functions.

[0176] In order to more intuitively illustrate the working principle and performance of the present invention, a specific application example is given below.

[0177] In this example, the system of the present invention is used to predict the physical properties of a petrochemical mixture. This mixture consists of n-hexane (40 wt%), cyclohexane (30 wt%), and toluene (30 wt%). The density, viscosity, surface tension, and octane number of the mixture need to be predicted over the temperature range of 293.15–373.15 K and pressure of 0.1–5.0 MPa.

[0178] First, the Data Processing and Management module collects physical property data for the three components, including pure substances and some mixtures, from the NIST ThermoDataEngine, the DIPPR database, and public literature. After cleaning and quality control, these data are generated into a standardized dataset. The Multiscale Feature Characterization module then constructs feature representations for the three components at four levels: quantum, molecular, mesoscopic, and macroscopic, and integrates these features through an attention mechanism. The Physical Constraint Modeling module then embeds thermodynamic laws and equation of state constraints into the model to ensure thermodynamic consistency of the prediction results. The Component Interaction Modeling module accurately captures the interactions between the three components, including binary interactions (n-hexane-cyclohexane, n-hexane-toluene, and cyclohexane-toluene) and ternary synergistic effects. The Uncertainty Quantification module assesses prediction uncertainty using Bayesian neural networks and Monte Carlo dropout methods. Finally, the Property Prediction and Analysis module, based on this processing, predicts the desired properties and analyzes structure-property relationships.

[0179] The predicted results show that the mixture's density at 298.15K and 0.1MPa is 783.2±1.5kg / m³, its viscosity is 0.52±0.02mPa·s, its surface tension is 24.3±0.5mN / m, and its octane number is 72.5±1.2. Compared with the experimental measurements, the relative error of the density prediction is 0.3%, the relative error of the viscosity prediction is 2.5%, the relative error of the surface tension prediction is 1.8%, and the absolute error of the octane number prediction is 0.8 units. These results significantly outperform traditional methods (such as ideal mixing rules or empirical correlations), which typically have errors 2-5 times greater than those of this system.

[0180] Furthermore, structure-property correlation analysis revealed that the benzene ring structure of toluene is the most important factor influencing the viscosity of the mixture (contributing approximately 45%), followed by the cyclohexane ring structure (contributing approximately 30%) and the chain length of n-hexane (contributing approximately 15%). The synergistic effect between the three components contributes approximately 10%, explaining why simple mixing rules are unable to accurately predict the viscosity of the mixture. This in-depth structure-property correlation analysis provides important guidance for optimizing mixture formulations.

Claims

1. A physical information-constrained petrochemical material mixture modeling and physical property calculation system, characterized by: include: Data processing and management module, used to: collect multi-source data of petrochemical materials; Cleaning and quality control the multi-source data; integrating the multi-source data into standard format data; A multi-scale feature characterization module is in communication with the data processing and management module and is used to: receive the standard format data; construct a multi-level petrochemical material feature characterization from quantum scale to macro scale; To achieve the fusion of features at different scales, the multi-scale feature characterization module includes a quantum chemistry feature characterization unit, a molecular structure feature characterization unit, a mesoscopic scale feature characterization unit, a macroscopic thermodynamic feature characterization unit, and a multi-scale feature fusion unit; a physical constraint modeling module, in communication with the multi-scale characterization module, configured to: receive the multi-level petrochemical material characterization; embed thermodynamic laws and physical law constraints into the model structure; and ensure that the physical property prediction results meet thermodynamic consistency requirements; A component interaction modeling module is in communication with the multi-scale feature characterization module and the physical constraint modeling module, and is used to: accurately capture the interactions between different components in a petrochemical material mixture; predict non-ideal mixing effects between components; and build a component interaction knowledge base; An uncertainty quantification module, in communication with the physical constraint modeling module and the component interaction modeling module, is used to: evaluate the reliability of the physical property prediction results; quantify the sources and magnitude of the uncertainty in the prediction; and generate confidence intervals for the prediction results; The physical property prediction and analysis module is in communication with the physical constraint modeling module, the component interaction modeling module, and the uncertainty quantification module, and is used to: predict various physical and chemical properties of petrochemical materials based on the outputs of the above modules; analyze the relationship between molecular structure and physical properties; and provide visualization and interpretation functions.

2. The system according to claim 1, wherein: The data processing and management module includes: a multi-source data acquisition unit, which is used to collect petrochemical material data from experimental databases, theoretical calculation results and industrial process data; a data cleaning and quality control unit, which is communicated with the multi-source data acquisition unit and is used to: detect and process outliers; verify the thermodynamic consistency of data; and perform quality scoring on data; a data integration and standardization unit, which is communicated with the data cleaning and quality control unit and is used to: unify the physical property data format; standardize molecular representation; and convert physical property units; a knowledge graph construction unit, which is communicated with the data integration and standardization unit and is used to: extract the relationship between material entities and physical properties; construct material-property-condition triples; and realize knowledge reasoning and completion.

3. The system according to claim 1, wherein: The multi-scale feature characterization module includes: a quantum chemical feature characterization unit, which is used to extract molecular electronic structure features; calculate molecular dipole moments and multipole moments; characterize intermolecular interactions; a molecular structure feature characterization unit, which is communicated with the quantum chemical feature characterization unit and is used to extract molecular topological structure features; construct molecular geometric structure features; identify and characterize functional groups; generate molecular fingerprints; a mesoscopic scale feature characterization unit, which is communicated with the molecular structure feature characterization unit and is used to characterize molecular aggregate features; extract phase behavior features; construct statistical thermodynamic features; a macroscopic thermodynamic feature characterization unit, which is communicated with the mesoscopic scale feature characterization unit and is used to construct state equation parameter features; extract thermodynamic property features; characterize transport property features; and a multi-scale feature fusion unit, which is communicated with each of the above feature characterization units and is used to achieve effective fusion and weight optimization of features of different scales.

4. The system according to claim 1, wherein: The physical constraint modeling module includes: a thermodynamic constraint embedding unit, which is used to implement the energy conservation constraint of the first law of thermodynamics; implement the entropy increase principle constraint of the second law of thermodynamics; embed phase equilibrium condition constraints; verify thermodynamic consistency relations; a state equation constraint unit, which is communicated with the thermodynamic constraint embedding unit and is used to embed cubic state equation constraints; embed SAFT series advanced state equation constraints; implement mixing rule constraints; perform special condition state equation corrections; a transport process constraint unit, which is communicated with the thermodynamic constraint embedding unit and is used to embed viscosity model constraints; embed heat conduction constraints; embed diffusion coefficient constraints; embed surface property constraints; a physical constraint optimization integration unit, which is communicated with the above-mentioned constraint units and is used to optimize the weights of each physical constraint; combine soft and hard constraint mechanisms; coordinate multi-scale constraint relationships; and evaluate the constraint embedding effect.

5. The system according to claim 1, wherein: The component interaction modeling module includes: a binary component interaction unit, which is used to calculate binary interaction parameters; predict binary mixing excess properties; predict binary phase diagrams; simulate special binary interaction effects; a multi-component interaction unit, which communicates with the binary component interaction unit and is used to identify high-order interaction terms; optimize multi-component mixing rules; predict multi-component phase behavior; simulate multi-component dynamic interactions; a component interaction characterization network unit, which communicates with the multi-component interaction unit and is used to construct a component graph representation learning framework; realize concentration-aware interaction networks; construct condition-sensitive interaction models; capture long-range interactions; a non-ideal mixing effect unit, which communicates with the component interaction characterization network unit and is used to realize activity coefficient models; comprehensively model excess properties; handle special non-ideal systems; and predict phase separation phenomena.

6. The system according to claim 1, wherein: The uncertainty quantification module includes: a prediction uncertainty source analysis unit, which is used to evaluate data uncertainty; analyze model uncertainty; quantify physical approximation uncertainty; evaluate prediction condition extrapolation uncertainty; a Bayesian uncertainty quantification unit, which is communicated with the prediction uncertainty source analysis unit and is used to implement a Bayesian neural network; implement Monte Carlo Dropout; construct Bayesian ensemble learning; implement deep Gaussian process; an uncertainty propagation and aggregation unit, which is communicated with the Bayesian uncertainty quantification unit and is used to analyze the uncertainty propagation chain; aggregate multi-source uncertainty; analyze condition dependency uncertainty; handle extreme condition uncertainty; an uncertainty calibration and evaluation unit, which is communicated with the uncertainty propagation and aggregation unit and is used to calibrate uncertainty estimation; evaluate prediction reliability; visualize uncertainty distribution; and detect abnormal predictions.

7. The system according to claim 1, wherein: The physical property prediction and analysis module includes: a thermodynamic property prediction unit, which is used to predict state properties, including density, specific heat capacity, enthalpy and entropy; predict phase equilibrium properties, including vapor pressure and critical parameters; calculate thermodynamic derivative properties; predict mixing properties; a transport property prediction unit, which is communicated with the thermodynamic property prediction unit and is used to predict viscosity; predict thermal conductivity properties; calculate diffusion coefficient; predict surface properties; a combustion property prediction unit, which is communicated with the thermodynamic property prediction unit and is used to predict fuel evaluation index, including octane number and cetane number; predict fire safety properties, including flash point and ignition point; calculate combustion thermodynamic properties; predict combustion kinetic parameters; a structure-property correlation analysis unit, which is communicated with each of the above prediction units and is used to identify key structural features; predict structural modification effects; analyze component interaction mechanisms; and analyze driving factors of physical property changes.

8. The system according to claim 1, wherein: The system also includes: a model training and optimization module, which is in communication with the multi-scale feature characterization module, the physical constraint modeling module, the component interaction modeling module and the uncertainty quantification module, and is used to implement multi-task joint learning; optimize physical constraint implementation; optimize model architecture; optimize hyperparameters; evaluate model performance and select the optimal model.

9. The system according to claim 1, wherein: The system also includes: a system integration and application module, which is in communication with the physical property prediction and analysis module and is used to provide a user interaction interface; achieve integration with industrial systems; perform data exchange and standardization; support system deployment and expansion; and adapt to different application scenarios.

10. A method for calculating physical properties of petrochemical material mixture modeling constrained by physical information, based on the system according to any one of claims 1 to 9, characterized in that: include: Collect multi-source data of petrochemical materials, and perform cleaning, quality control, and standardization on the multi-source data to obtain data in a standard format; Based on the standard format data, a multi-level petrochemical material characterization from the quantum scale to the macroscale is constructed, and the characteristics of different scales are integrated. Thermodynamic laws and physical law constraints are embedded in the model structure to ensure that the physical property prediction results meet the requirements of thermodynamic consistency. The complex interactions between different components in the petrochemical material mixture are accurately captured, the non-ideal mixing effects between components are predicted, and a knowledge base of component interactions is constructed. Evaluate the reliability of physical property prediction results, quantify the sources and magnitude of uncertainty in the predictions, and generate confidence intervals for the prediction results; Based on the above steps, various physical and chemical properties of petrochemical materials are predicted, the relationship between molecular structure and physical properties is analyzed, and visualization and interpretation functions are provided.

Images