An AI-based chemical reaction experiment simulation system and method

Abstract

The present invention relates to the technical field of chemical reaction experiment simulation, and discloses an AI-based chemical reaction experiment simulation system and method. The system includes a reaction mechanism model and a dynamic simulation model. The former integrates and processes the initial experimental data, and the latter dynamically suppresses the disturbance of the real-time reaction data stream and inputs it into the path generation layer to simulate the reaction trajectory and generate state instructions. The path generation layer contains a data preprocessing module and a dynamic path generation module, which realizes the identification of key reaction nodes, compensation field data generation and state instruction optimization through energy level correlation processing, dynamic matching modeling, multi-dimensional parameter fusion, etc. The present invention improves the accuracy of multi-dimensional data integration, dynamic disturbance processing and trajectory simulation, realizes precise control of reaction state and high-risk warning, reduces experimental costs and risks, and is suitable for reaction simulation in chemical research and industrial production.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical reaction experiment simulation, and specifically to an AI-based chemical reaction experiment simulation system and method. Background Art

[0002] In chemical research and industrial production, the design, optimization, and safety assessment of chemical reaction experiments have always been core challenges. Traditional experimental methods rely on manual trial and error, which have significant drawbacks such as high cost, long cycle time, and high risk. Especially for complex chemical reaction systems, it is difficult to accurately capture the dynamic changes and multi-parameter coupling effects during the reaction process. With the development of computer technology and artificial intelligence, simulation technology has gradually become an important means to replace traditional experiments, but existing technologies still face many bottlenecks:

[0003] Existing simulation systems are often only capable of processing a single type of experimental data, such as modeling reactant properties or reaction conditions, and struggle to integrate multi-dimensional data such as reactant properties, reaction conditions, and catalyst distribution. For example, traditional models cannot effectively correlate molecular structural topology parameters with chemical bond energy parameters, resulting in an incomplete analysis of reaction mechanisms and an inability to accurately predict reaction pathways and product distributions.

[0004] Chemical reactions are often accompanied by real-time fluctuations in environmental parameters such as temperature, pressure, and concentration. Existing systems lack efficient dynamic disturbance suppression algorithms, making it difficult to maintain signal stability during data transmission and processing. For example, when the reaction system is subject to external interference, the real-time data stream may experience noise or abnormal fluctuations, resulting in significant deviations between the simulation results and the actual reaction trajectory.

[0005] Traditional pathway generation models are often based on fixed rules or simple fits of historical data, lacking the ability to dynamically respond to real-time environmental parameters. For example, when simulating complex catalytic reactions, they are unable to adjust the reaction pathway based on real-time changes in catalyst distribution. This results in delayed capture of key reaction nodes, impacting the accuracy of simulation results.

[0006] Chemical reactions involve the synergistic effects of multiple parameters, including reactants, conditions, and catalysts. Existing systems struggle to establish dynamic matching relationships between these parameters. For example, when processing path coupling feature data from different eigenvectors, it's impossible to effectively calculate the deviation coefficients between parameters and generate compensation field data, resulting in inaccurate control of the reaction state.

[0007] Existing systems lack real-time anomaly identification and early warning mechanisms for high-risk reaction states (such as intense heat release and explosion risks), making it impossible to generate timely intervention instructions before the reaction gets out of control. For example, when a reaction system experiences abnormal energy fluctuations, traditional systems may be unable to quickly locate high-risk nodes and adjust reaction conditions, resulting in a high safety risk. Summary of the Invention

[0008] The purpose of the present invention is to provide an AI-based chemical reaction experiment simulation system and method to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an AI-based chemical reaction experiment simulation system, the system comprising:

[0010] A reaction mechanism model for performing multi-dimensional parameter integration processing on initial experimental data, wherein the initial experimental data includes a first eigenvector corresponding to the reactant physical property characteristic data, a second correlation matrix corresponding to the reaction condition characteristic data, and a third mapping sequence corresponding to the catalyst distribution characteristic data. The reactant physical property characteristic data includes a first distribution label generated by molecular structure topology parameters and a second quantitative label generated by chemical bond energy parameters.

[0011] A dynamic simulation model is used to perform dynamic disturbance suppression processing on the real-time reaction data stream and input it into the path generation layer to simulate the reaction trajectory, and generate reaction state instructions according to the output results of the path generation layer;

[0012] The path generation layer includes a data preprocessing module and a dynamic path generation module, wherein the data preprocessing module is used to split the original reaction data stream into energy intervals and filter invalid features, and the dynamic path generation module is obtained through collaborative training based on multiple types of historical reaction data and real-time environmental parameters; the dynamic path generation module includes a path parsing layer, a path correction layer, and a state triggering layer connected in sequence.

[0013] Preferably, the path parsing layer is used to perform energy level correlation processing on different eigenvectors in the original reaction data stream to generate path coupling characteristic data; the path correction layer is used to model the dynamic matching relationship between the path coupling characteristic data corresponding to each eigenvector to generate compensation field data; the state triggering layer is used to perform multi-dimensional parameter fusion based on the compensation field data and the path coupling characteristic data to generate reaction state instructions.

[0014] Preferably, the modeling of the dynamic matching relationship between the path coupling characteristic data corresponding to each characteristic vector to generate compensation field data includes:

[0015] A path segmentation algorithm is used to identify the key reaction nodes in the path coupling characteristic data, and a correction compensation sequence corresponding to each characteristic vector is determined based on the environmental parameter type corresponding to each key reaction node;

[0016] The deviation coefficient between the reaction nodes of the same parameter in the correction compensation sequence corresponding to any two eigenvectors is calculated, and the compensation field data between the any two eigenvectors is generated based on the deviation coefficient.

[0017] Preferably, the calculation of the deviation coefficient between reaction nodes with the same parameters in the correction compensation sequence corresponding to any two eigenvectors includes:

[0018] When the number of reaction nodes in the correction compensation sequence corresponding to any two eigenvectors is inconsistent, virtual reaction point interpolation is performed based on the environmental parameters corresponding to the terminal reaction node among the reaction nodes with fewer reaction nodes, and the deviation coefficient between the reaction nodes with the same parameters is calculated based on the interpolated data.

[0019] Preferably, the data preprocessing module is specifically used to:

[0020] Performing isogradient division on the first eigenvector, the second correlation matrix, and the third mapping sequence according to a preset energy interval to generate standardized reaction data, standardized condition data, and standardized catalytic data;

[0021] A dynamic parameter matching method is used to align the standardized reaction data and the standardized condition data in real time, and a fixed energy window mechanism is used to perform steady-state optimization on the standardized catalytic data, outputting a first calibration vector, a second calibration matrix, and a third calibration sequence; wherein the first calibration vector includes a calibrated first distribution label and a calibrated second quantization label.

[0022] Preferably, the data preprocessing module is further used for:

[0023] Calculating a path deviation coefficient between the calibrated first distribution label and the calibrated second quantization label within a historical reaction period;

[0024] Predicting an expected distribution value of the calibrated second quantized label in the real-time response phase according to the path deviation coefficient and the parameter value of the calibrated first distribution label in the real-time response phase;

[0025] Target correction compensation data is generated based on the calibrated second quantization label and its expected distribution value, and a feature vector corresponding to the target correction compensation data is used as a first calibration vector.

[0026] Preferably, the path correction layer specifically includes:

[0027] a path tracing unit, configured to perform reaction path tracing on each characteristic vector in the path coupling characteristic data, so as to extract a corresponding reaction propagation chain from each characteristic vector;

[0028] The parameter matching unit is used to perform energy level superposition on the reaction propagation chain extracted by each eigenvector and the corresponding path coupling characteristic data to generate compensation field data.

[0029] Preferably, the path correction layer further includes:

[0030] The noise elimination unit is used to perform phase lag effect elimination processing on the compensation field data.

[0031] Preferably, the state trigger layer specifically includes:

[0032] A multi-dimensional parameter coordination unit, comprising a plurality of state decision nodes, each state decision node being connected to each characteristic vector in the compensation field data and the path coupling characteristic data through configuration parameters;

[0033] a dynamic gradient optimization unit, configured to iteratively optimize the configuration parameters using a dynamic gradient adjustment algorithm to minimize the error between the reaction state command and the actual reaction trajectory;

[0034] The abnormal trajectory identification unit is used to locate high-risk reactions based on the compensation field data and the path coupling characteristic data, and generate reaction state instructions.

[0035] Preferably, the present invention also includes an AI-based chemical reaction experiment simulation method, which is applied to the above-mentioned AI-based chemical reaction experiment simulation system, and the method includes the following steps:

[0036] Performing multi-dimensional parameter integration processing on initial experimental data using a reaction mechanism model, wherein the initial experimental data includes a first eigenvector corresponding to reactant physical property characteristic data, a second correlation matrix corresponding to reaction condition characteristic data, and a third mapping sequence corresponding to catalyst distribution characteristic data, wherein the reactant physical property characteristic data includes a first distribution label generated by molecular structure topology parameters and a second quantitative label generated by chemical bond energy parameters;

[0037] The real-time reaction data stream is subjected to dynamic disturbance suppression processing through a dynamic simulation model, and the processed data stream is input into the path generation layer for reaction trajectory simulation;

[0038] The data preprocessing module in the path generation layer is used to split the original reaction data stream into energy intervals and filter invalid features;

[0039] The reaction trajectory simulation is performed using a dynamic path generation module obtained by collaborative training of multiple types of historical reaction data and real-time environmental parameters. The dynamic path generation module includes a path parsing layer, a path correction layer, and a state triggering layer connected in sequence.

[0040] According to the output results of the path generation layer, the dynamic simulation model generates reaction state instructions.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The system uses a reaction mechanism model to perform multi-dimensional parameter integration on the initial experimental data, organically integrating the reactant physical property characteristic data (such as the first distribution label generated by molecular structure topology parameters and the second quantitative label generated by chemical bond energy parameters), reaction condition characteristic data (the second correlation matrix), and catalyst distribution characteristic data (the third mapping sequence). This integration method breaks through the limitations of traditional models' single-dimensional data processing and can analyze reaction mechanisms from multiple levels, including molecular structure, energy changes, conditional parameters, and catalytic distribution, providing a more comprehensive data source for accurately predicting reaction pathways. For example, by correlating molecular topology and bond energy parameters, reaction active sites and energy transfer pathways can be more accurately determined, thereby improving the simulation accuracy of the reaction thermodynamic and kinetic characteristics.

[0043] The dynamic simulation model effectively eliminates noise interference caused by environmental parameter fluctuations by dynamically suppressing real-time reaction data streams, ensuring data stability. Furthermore, the dynamic path generation module of the path generation layer is collaboratively trained based on multiple types of historical reaction data and real-time environmental parameters, dynamically adjusting the reaction trajectory simulation strategy based on real-time changing conditions (such as temperature, pressure, and catalyst concentration). For example, in the data preprocessing module, key reaction features can be rapidly extracted through energy interval splitting and invalid feature filtering. The path parsing layer performs energy level correlation processing on different eigenvectors to generate path coupling feature data, enabling the system to capture the dynamic correlations between various parameters during the reaction process. The path correction layer further corrects the simulation path by modeling the dynamic matching relationship between eigenvectors and generating compensation field data, reducing errors caused by parameter deviations. This dynamic response mechanism significantly enhances the system's real-time simulation capabilities for complex reaction systems, making the simulation results more closely aligned with the actual reaction process.

[0044] The path correction layer uses a path segmentation algorithm to identify key reaction nodes, determines the correction compensation sequence based on the type of environmental parameters, and calculates the deviation coefficient between the characteristic vectors to generate compensation field data. This mechanism can effectively compensate for simulation deviations caused by dynamic changes in multiple parameters and achieve precise control of the reaction state. For example, when there is a mismatch between the reactant distribution and the reaction condition parameters, the system can dynamically adjust the weights of each parameter through the compensation field data, optimize the reaction path, and ensure that the reaction proceeds in the expected direction. The state trigger layer, through the synergistic effect of the multi-dimensional parameter coordination unit, the dynamic gradient optimization unit, and the abnormal trajectory identification unit, can not only generate accurate reaction state instructions, but also minimize simulation errors by iteratively optimizing the configuration parameters, and locate high-risk reaction nodes in real time, issuing warnings or intervention instructions in advance, significantly improving the safety and controllability of the reaction process.

[0045] The data preprocessing module performs isogradient division and standardization on multi-dimensional data through preset energy intervals, and uses a dynamic parameter matching method to align reaction data and condition data in real time, while optimizing catalytic data through a fixed energy window mechanism. This preprocessing mechanism effectively improves the standardization and consistency of the data and reduces calculation errors caused by inconsistent data formats or energy fluctuations. For example, by calculating the path offset coefficient and predicting the expected distribution value of the quantitative label, the system can generate target correction compensation data, further calibrate the input vector, and enhance the system's adaptability and robustness to different reaction systems. In addition, the noise elimination unit eliminates the phase lag effect of the compensation field data, ensuring the real-time transmission and processing of data and avoiding simulation distortion caused by signal delays.

[0046] The present invention uses AI technology to achieve full-process simulation of chemical reaction experiments, without relying on a large number of actual experiments, significantly reducing R&D costs and time cycles. At the same time, the system's real-time abnormality recognition and early warning functions can effectively prevent the occurrence of high-risk reactions and reduce the risk of casualties and equipment damage caused by experimental accidents. For example, when simulating reactions with explosion risks, the system can monitor abnormal energy fluctuations in real time, quickly locate dangerous nodes and adjust reaction conditions to control risks in the bud. This intelligent simulation method provides a safe and efficient experimental design platform for chemical research and industrial production, with significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a working principle diagram of the AI-based chemical reaction experiment simulation system and method described in the present invention;

[0048] Figure 2 A diagram of the working principle of the dynamic path generation module;

[0049] Figure 3 Flowchart for path correction layer compensation field generation;

[0050] Figure 4 Flowchart for calibration of data preprocessing module;

[0051] Figure 5 It is the fusion diagram of multi-dimensional parameters of the state trigger layer. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] See also Figure 1-Figure 5 The present invention provides a specific embodiment of an AI-based chemical reaction experiment simulation system, which includes a reaction mechanism model and a dynamic simulation model. The following is a detailed description of the specific structure and function:

[0054] The reaction mechanism model is used to perform multi-dimensional parameter integration processing on the initial experimental data. The initial experimental data includes a first eigenvector corresponding to the reactant physical property characteristic data, a second correlation matrix corresponding to the reaction condition characteristic data, and a third mapping sequence corresponding to the catalyst distribution characteristic data. The reactant physical property characteristic data includes a first distribution label generated by molecular structure topology parameters and a second quantitative label generated by chemical bond energy parameters. In specific implementation, the reaction mechanism model constructs a multi-dimensional data fusion algorithm to perform structured processing on the first eigenvector, second correlation matrix, and third mapping sequence. For example, the molecular structure topology parameters (such as atomic connection order, bond length, bond angle, etc.) are converted into a first distribution label with spatial distribution characteristics, and the chemical bond energy parameters (such as bond energy value, energy level transition probability, etc.) are converted into a quantifiable second quantitative label. Through matrix operations and sequence analysis, a comprehensive characterization of the reactant physical properties, reaction conditions, and catalyst distribution is achieved.

[0055] The dynamic simulation model performs dynamic disturbance suppression on the real-time reaction data stream and feeds it into the path generation layer for reaction trajectory simulation. The output of the path generation layer generates reaction state instructions. The path generation layer consists of a data preprocessing module and a dynamic path generation module. The data preprocessing module is responsible for energy binning and invalid feature filtering on the raw reaction data stream. Specifically, the raw reaction data stream may contain multi-dimensional data such as reactant concentration changes, temperature fluctuations, and pressure changes. The data preprocessing module divides the data stream into bins using preset energy thresholds, removing invalid features with abnormal energy values ​​or minimal impact on the reaction progress, thereby improving the accuracy and efficiency of subsequent simulations. The dynamic path generation module is trained collaboratively based on multiple types of historical reaction data and real-time environmental parameters. It consists of a sequentially connected path parsing layer, a path correction layer, and a state triggering layer. During training, the module collects historical data from different chemical reactions (such as acid-base neutralization reactions and redox reactions) and combines it with real-time environmental parameters (such as temperature, pressure, and catalyst concentration). Deep learning algorithms (such as recurrent neural networks and graph neural networks) are then used to train the module, enabling it to learn reaction path patterns under different reaction conditions.

[0056] Example 1:

[0057] In the AI-based chemical reaction experiment simulation system, the path parsing layer performs energy level correlation processing on different eigenvectors in the original reaction data stream to generate path coupling characteristic data. The first eigenvector, second correlation matrix and third mapping sequence contained in the original reaction data stream correspond to the reactant physical property characteristic data, reaction condition characteristic data and catalyst distribution characteristic data, respectively. The path parsing layer first processes the first eigenvector, which contains a first distribution label generated by the molecular structure topology parameters and a second quantitative label generated by the chemical bond energy parameters. The molecular structure topology parameters involve spatial structural information such as atomic connection mode, bond length, bond angle, etc., while the chemical bond energy parameters are related to the stability and reactivity of chemical bonds within the molecule.

[0058] The path parsing layer constructs an energy level correlation model to analyze the energy transfer relationship between the molecular structure topology parameters and chemical bond energy parameters in the first eigenvector. For example, for reactant molecules with specific spatial structures, their chemical bond energy parameters determine the reactivity and energy changes of the molecules under specific conditions. By calculating the projection and correlation of the molecular structure topology parameters and chemical bond energy parameters in energy space, the path parsing layer can identify characteristic parameter combinations that have a critical impact on the reaction process. When processing the second correlation matrix, the path parsing layer analyzes the energy correlation between the parameters in the reaction condition characteristic data (such as temperature, pressure, concentration, etc.) and the physical property characteristic data of the reactants. Different reaction conditions affect the energy state of the reactant molecules, thereby changing the reaction path and rate. The path parsing layer couples the reaction condition characteristic data with the physical property characteristic data by establishing a multidimensional energy correlation model to analyze and identify synergistic interaction patterns between the reaction conditions and the reactants.

[0059] For the third mapping sequence, the path parsing layer analyzes the energy correlations between catalyst distribution feature data and reactant property feature data and reaction condition feature data. Catalysts accelerate reactions by lowering the activation energy of reactions, and their distribution characteristics influence the selectivity and efficiency of reactions. The path parsing layer constructs a three-dimensional energy correlation model of catalyst, reactants, and reaction conditions to analyze the impact of catalyst distribution within the reaction system on the reaction energy state, thereby determining the catalyst's mechanism of action at different reaction stages. In generating path coupling feature data, the path parsing layer employs a multiscale analysis approach, comprehensively considering energy transfer and interactions at different scales, from the molecular level to the macroscopic reaction system. For example, at the molecular level, it analyzes the energy changes during the breaking and forming of chemical bonds within reactant molecules; at the macroscopic reaction system level, it analyzes the influence of macroscopic parameters such as temperature and pressure on the reaction energy distribution. By integrating energy information at different scales, the path coupling feature data generated by the path parsing layer comprehensively reflects the energy correlations between the various eigenvectors in the reaction system, providing an accurate data foundation for subsequent path correction and state triggering.

[0060] The path correction layer models the dynamic matching relationship between the path coupling feature data corresponding to each eigenvector and generates compensation field data. This layer first uses a path segmentation algorithm to identify key reaction nodes within the path coupling feature data. Key reaction nodes are nodes with significant energy changes or material transformations during the reaction process. These nodes typically correspond to rate-determining steps or the formation of key intermediates. The path segmentation algorithm divides the path coupling feature data into multiple reaction stages based on metrics such as energy change rate and material transformation rate, with each stage bounded by a key node. Based on the environmental parameter type corresponding to each key reaction node, the path correction layer determines the corresponding compensation sequence for each eigenvector. Environmental parameter types include temperature, pressure, and catalyst concentration, and different environmental parameters have different effects on the reaction process. For each key reaction node, the path correction layer analyzes the difference between the parameter values ​​of each eigenvector at that node and the ideal parameter values ​​and generates a corresponding compensation sequence. For example, at temperature-sensitive key reaction nodes, the reactant property eigenvectors and reaction condition eigenvectors need to be compensated based on temperature changes to ensure the accuracy of the reaction path.

[0061] The deviation coefficient is calculated between reaction nodes with identical parameters in the corrected compensation sequences corresponding to any two eigenvectors. When the number of reaction nodes in two corrected compensation sequences differs, the path correction layer interpolates virtual reaction points based on the environmental parameters corresponding to the terminal reaction node in the sequence with the fewer reaction nodes. Virtual reaction point interpolation is a data processing method that inserts virtual nodes into the sequence with the fewer reaction nodes to make the number of nodes in the two sequences consistent, facilitating the calculation of the deviation coefficient. The interpolation process is based on the environmental parameters and reaction trends of the terminal reaction nodes, ensuring that the inserted virtual nodes have physical meaning. Based on the calculated deviation coefficient, the path correction layer generates compensation field data between any two eigenvectors. This compensation field data reflects the dynamic matching differences between the eigenvectors and is used to correct deviations in the reaction path. In this way, the path correction layer can precisely adjust the reaction path and improve the accuracy of reaction simulations.

[0062] The state trigger layer fuses multidimensional parameters based on the compensation field data and path coupling characteristic data to generate reaction state instructions. This layer constructs a multidimensional parameter fusion model to comprehensively analyze the various dimensional parameters in the compensation field data and path coupling characteristic data. Multidimensional parameters include energy parameters, substance concentration parameters, reaction rate parameters, etc., which describe the state of the reaction system from different perspectives. The state trigger layer converts these multidimensional parameters into instructions that can indicate the reaction state through weighted summation and neural network mapping. For example, by performing a weighted summation of energy parameters, substance concentration parameters, and reaction rate parameters, a comprehensive reaction state indicator is obtained, and the corresponding reaction state instruction is generated based on this indicator.

[0063] The state triggering layer comprises multiple state decision nodes, each corresponding to a specific reaction state or control strategy. These nodes are connected to eigenvectors in the compensation field data and path coupling characteristic data via configuration parameters. These configuration parameters determine each state decision node's sensitivity to and response to different eigenvectors. For example, a state decision node corresponding to a temperature control strategy has configuration parameters that determine the node's weight and threshold for temperature-related eigenvectors. When the parameter value of a temperature-related eigenvector exceeds the threshold, the node is activated and generates the corresponding temperature control command. The state triggering layer optimizes the decision-making capabilities of the state decision nodes by dynamically adjusting the configuration parameters. This dynamic adjustment process, based on a feedback mechanism, continuously adjusts the configuration parameters based on the discrepancy between the reaction state command and the actual reaction trajectory, ensuring that the state decision node more accurately reflects the actual reaction state. For example, during simulation, the error between the reaction state command and the actual reaction trajectory is monitored in real time. The configuration parameters are adjusted based on the direction and magnitude of the error, gradually improving the decision-making accuracy of the state decision node.

[0064] The reaction state instructions generated by the state trigger layer include instructions for adjusting reaction rate, adding catalysts, and adjusting temperature. These instructions are dynamically generated based on the real-time state of the reaction system and are used to control the progress of the chemical reaction experiment. For example, if the reaction rate is too slow, an instruction is generated to increase the catalyst concentration or raise the temperature; if the reaction rate is too fast, an instruction is generated to lower the temperature or reduce the catalyst dosage. In this way, the state trigger layer enables real-time intelligent control of chemical reaction experiments, improving the accuracy and controllability of reaction simulations. When generating reaction state instructions, the state trigger layer also considers the stability and safety of the reaction system. For example, when generating temperature adjustment instructions, the impact of temperature changes on the stability of the reaction system is considered to avoid runaway reactions or unsafe conditions caused by excessively high or low temperatures. By comprehensively considering these factors, the reaction state instructions generated by the state trigger layer can ensure the stability and safety of the reaction system while maintaining reaction efficiency.

[0065] Example 2:

[0066] In the AI-based chemical reaction experiment simulation system, the data preprocessing module undertakes the important task of standardizing, calibrating and optimizing the raw data. The module first performs isogradient division on the first eigenvector, the second correlation matrix and the third mapping sequence according to the preset energy interval. The setting of the preset energy interval is based on the energy characteristics of different chemical reactions. For example, different energy threshold ranges are set for endothermic reactions and exothermic reactions. During the isogradient division process, the module divides the reactant physical property characteristic data in the first eigenvector evenly into each preset interval according to the energy size to generate standardized reaction data. Similarly, the reaction condition characteristic data in the second correlation matrix and the catalyst distribution characteristic data in the third mapping sequence are also divided isogradiently to generate standardized condition data and standardized catalytic data, respectively. This division method ensures that different types of data are comparable in energy scale, laying the foundation for subsequent processing and analysis.

[0067] After completing data standardization, the data preprocessing module uses a dynamic parameter matching method to align the standardized reaction data and standardized condition data in real time. This method generates adaptive matching parameters based on the reactant distribution characteristics of real-time environmental parameters. Real-time environmental parameters include temperature, pressure, and reactant concentration, which are acquired in real time by sensors. The reactant distribution characteristics are obtained by analyzing sensor data, such as using concentration sensors to monitor the spatial distribution of reactants within the reaction vessel. Based on these distribution characteristics, the system generates adaptive matching parameters that dynamically adjust the data alignment based on changes in the reactant distribution. An energy gradient segmentation mechanism is used to perform multi-dimensional calibration on the standardized reaction data. This energy gradient segmentation mechanism divides the reaction process into multiple stages based on the energy gradient, each with different energy and reaction rate characteristics. Within each stage, the system uses calibration parameters specific to different dimensions (such as temperature, pressure, and concentration) for calibration. For example, during the high-temperature stage, the system prioritizes calibration of temperature-related parameters to ensure real-time alignment of the reaction data and reaction condition data for that stage.

[0068] For standardized catalytic data, the data preprocessing module uses a fixed energy window mechanism for steady-state optimization. The fixed energy window mechanism refers to setting a fixed energy window range during the catalytic data processing process, and smoothing and steady-state optimizing the catalytic data within the window. This mechanism can eliminate random fluctuations and noise in the catalytic data and improve the stability and reliability of the data. Through the steady-state optimization of the catalytic data, the system can more accurately reflect the mechanism of action of the catalyst in the reaction and provide more reliable data support for subsequent reaction simulations. After the above processing, the data preprocessing module outputs the first calibration vector, the second calibration matrix and the third calibration sequence. The first calibration vector contains the calibrated first distribution label and the calibrated second quantitative label, which accurately reflect the physical properties of the reactants; the second calibration matrix and the third calibration sequence correspond to the calibrated reaction condition characteristic data and catalyst distribution characteristic data, respectively, providing an accurate data basis for subsequent reaction trajectory simulations.

[0069] The data preprocessing module also calculates the path deviation coefficient between the calibrated first distribution label and the calibrated second quantitative label over the historical reaction cycle. This path deviation coefficient is calculated based on the difference in their trajectories and correlation analysis over the historical cycle. During a chemical reaction, the molecular structure topology parameters (first distribution label) and chemical bond energy parameters (second quantitative label) of the reactants change as the reaction proceeds. These changes reflect the reaction's progress and mechanism. By analyzing the changes in the trajectories of these two labels over the historical reaction cycle, the system can calculate the path deviation coefficient between them. This coefficient is used to measure the degree of match between the molecular structure topology parameters and chemical bond energy parameters in the reactant's physical properties.

[0070] Based on the path offset coefficient and the parameter value of the calibrated first distribution label in the real-time reaction stage, the data preprocessing module predicts the expected distribution value of the calibrated second quantitative label in the real-time reaction stage. This prediction process is based on the established regression model or time series model, and uses the path offset coefficient in the historical data and the real-time parameter value of the first distribution label to obtain the expected distribution of the second quantitative label through model calculation. This prediction method makes full use of the information of historical reaction data, can more accurately estimate the parameter value of the real-time reaction stage, and improves the system's ability to predict the reaction process. Target correction compensation data is generated based on the calibrated second quantitative label and its expected distribution value. The target correction compensation data is used to correct the deviation in the reaction simulation process to ensure that the simulation results are closer to the actual reaction situation. The characteristic vector corresponding to the target correction compensation data is used as the first calibration vector, which further optimizes the accuracy of the reactant physical property characteristic data, enabling the system to more accurately simulate the chemical reaction process.

[0071] To calculate the path shift coefficient, the system first performs a time series analysis of the calibrated first distribution label and the calibrated second quantitative label over the historical reaction period. By analyzing the numerical changes of these two labels at different time points, their changing trends and periodic characteristics are extracted. The correlation between these two time series is then calculated to determine the degree of association between them. Based on this correlation and the differences in their trajectories, the system calculates the path shift coefficient, which reflects the degree of synergistic change between the two labels during the reaction process.

[0072] To predict the expected distribution value of the calibrated second quantitative label, the system employs regression or time series models trained and validated with extensive historical data. These models capture the inherent relationship between the first and second distribution labels, as well as their temporal variations. During real-time prediction, the model calculates the expected distribution value of the second quantitative label based on the current first distribution label parameter values ​​and the historical path deviation coefficients. This prediction method not only considers the current reaction state but also leverages historical reaction data, improving prediction accuracy and reliability.

[0073] In the process of generating target correction compensation data, the system first calculates the difference between the calibrated second quantitative label and its expected distribution value. This difference reflects the deviation between the change of the second quantitative label in the actual reaction process and the expected change. Then, based on this difference, the corresponding correction compensation data is generated, which contains the adjustment information for the second quantitative label. The characteristic vector corresponding to the target correction compensation data is used as the first calibration vector. In the subsequent reaction simulation process, the system can adjust the reaction trajectory according to these correction compensation data to make the simulation results more consistent with the actual reaction situation. In this way, the data preprocessing module realizes the dynamic optimization of the physical property characteristic data of the reactants, thereby improving the simulation accuracy and reliability of the system.

[0074] Example 3:

[0075] In the AI-based chemical reaction experiment simulation system, the path correction layer performs reaction path tracing on each eigenvector in the path coupling characteristic data through the path tracing unit. For the first eigenvector, that is, the reactant physical property characteristic data, the path tracing unit analyzes the change trajectory of the molecular structure topology parameters. The molecular structure topology parameters include spatial structure information such as atomic connection order, bond length, bond angle, etc. The changes in these parameters during the reaction process reflect the structural evolution path of the reactant molecules. By tracking the changes in these parameters, the system can extract the reaction propagation chain that reflects the reactant conversion process. For example, in an organic chemical reaction, the process of breaking certain chemical bonds of the reactant molecules and forming new chemical bonds will cause the molecular structure topology parameters to change. By analyzing these changes, the path tracing unit can determine the formation and conversion path of the reaction intermediates, thereby constructing a complete reaction propagation chain.

[0076] For the second correlation matrix, that is, the reaction condition characteristic data, the path tracing unit analyzes the change sequence of condition parameters such as temperature and pressure. Changes in reaction conditions will have a significant impact on the reaction rate and reaction path. By analyzing the change trajectory of these condition parameters, the system can trace the impact path of the reaction conditions on the reaction process. For example, an increase in temperature may accelerate certain reaction steps, while a change in pressure may affect the equilibrium position of the reaction. The path tracing unit extracts the reaction condition propagation chain by establishing an association model between the condition parameters and the reaction process. This chain describes how the reaction conditions affect the progress of the reaction over time.

[0077] For the third mapping sequence, that is, the catalyst distribution characteristic data, the path tracing unit analyzes the distribution changes of the catalyst in the reaction system. The catalyst accelerates the reaction process by reducing the activation energy of the reaction, and its distribution characteristics will affect the selectivity and efficiency of the reaction. The path tracing unit traces the catalytic action path of the catalyst on the reaction by tracking the concentration distribution, active site changes and other information of the catalyst at different reaction stages, and extracts the catalyst propagation chain. For example, in a multiphase catalytic reaction, the surface properties and active site distribution of the catalyst will change as the reaction proceeds. By analyzing these changes, the path tracing unit can determine the action mechanism of the catalyst at different reaction stages, providing a basis for subsequent reaction path corrections.

[0078] The parameter matching unit superimposes the energy levels of the reaction propagation chain extracted from each eigenvector with the corresponding path coupling characteristic data to generate compensation field data. During this energy level superposition, the parameter matching unit first compares and analyzes the energy parameters of each node in the reaction propagation chain with the energy parameters in the path coupling characteristic data. Energy parameters include reaction heat, activation energy, and Gibbs free energy, which directly influence the rate and direction of the reaction. Through this comparative analysis, the system determines the degree of energy matching between the reaction propagation chain and the path coupling characteristic data.

[0079] For parts with a higher degree of matching, the parameter matching unit directly superimposes the energy parameters in the reaction propagation chain with the corresponding parameters in the path coupling characteristic data. For example, at a certain reaction node, the activation energy parameters in the reaction propagation chain are similar to the activation energy parameters in the path coupling characteristic data. The system performs a weighted average of these two parameters to obtain a more accurate activation energy value. For parts with a lower degree of matching, the parameter matching unit performs further analysis and adjustment. The system will check the source of the difference in energy parameters between the reaction propagation chain and the path coupling characteristic data, which may be due to experimental errors, model simplifications, or changes in reaction conditions. For different reasons, the system adopts corresponding adjustment strategies. For example, if the difference is caused by experimental errors, the system will smooth the data; if the difference is caused by model simplification, the system will introduce a more complex model for correction.

[0080] After superimposing the energy parameters, the parameter matching unit also considers parameters from other dimensions, such as substance concentration and reaction rate. These parameters are interrelated with the energy parameters and jointly influence the reaction process. The system establishes a multidimensional parameter correlation model, integrating the energy parameters with parameters from other dimensions to generate the final compensation field data. This compensation field data reflects the matching differences between the eigenvectors in energy and other dimensions. This difference information is used for subsequent reaction path correction to improve the accuracy of the reaction simulation.

[0081] The noise elimination unit performs phase lag effect elimination processing on the compensation field data. During the transmission and processing of reaction data, factors such as sensor response time and data transmission delay may cause phase lag in the compensation field data. This phase lag will affect the system's accurate judgment of the reaction state and therefore needs to be eliminated. The noise elimination unit uses a digital filtering algorithm to process the compensation field data. The digital filtering algorithm is an algorithm based on signal processing theory. It can filter the signal according to the frequency characteristics of the signal and remove unnecessary frequency components. In the phase lag effect elimination process, the noise elimination unit first performs spectral analysis on the compensation field data to determine the frequency characteristics of the phase lag. Then, the corresponding filter is designed based on these frequency characteristics to filter the compensation field data.

[0082] When selecting a digital filtering algorithm, the noise cancellation unit considers various factors, such as filtering effectiveness, computational complexity, and real-time requirements. Common digital filtering algorithms for eliminating phase lag effects include finite impulse response (FIR) filters and infinite impulse response (IIR) filters. FIR filters have linear phase characteristics, ensuring that the signal's phase information is not distorted, making them suitable for applications with high phase requirements. IIR filters, on the other hand, offer higher filtering efficiency, achieving higher filtering performance with fewer coefficients, making them suitable for applications with limited computing resources. The noise cancellation unit selects the appropriate filtering algorithm based on the specific application requirements.

[0083] During the filtering process, the noise cancellation unit also considers the real-time requirements of the compensation field data. Because chemical reactions are dynamic, the system needs to obtain real-time information about the reaction state and make corresponding adjustments. Therefore, when designing the filter, the noise cancellation unit optimizes the algorithm's computational efficiency to ensure that the filtering of the compensation field data can be completed in a short period of time. The noise cancellation unit also performs a quality assessment on the filtered compensation field data to ensure that phase lag effects are eliminated without introducing new errors or distortion. In this way, the noise cancellation unit effectively improves the quality of the compensation field data and provides reliable data support for accurate correction of the reaction path.

[0084] Throughout the entire path correction process, the path correction layer achieves precise adjustment of the reaction path through the collaborative work of the path tracing unit, parameter matching unit, and noise elimination unit. The reaction propagation chain extracted by the path tracing unit provides basic data for the parameter matching unit. The compensation field data generated by the parameter matching unit through energy level superposition reflects the dynamic matching differences between each eigenvector, while the noise elimination unit ensures the accuracy and reliability of the compensation field data. These processing steps are interrelated and progressive, jointly improving the simulation accuracy and reliability of the AI-based chemical reaction experiment simulation system, enabling the system to more accurately predict reaction results and provide strong support for actual chemical reaction experiments.

[0085] Example 4:

[0086] In an AI-based chemical reaction experiment simulation system, the state triggering layer achieves intelligent analysis of reaction states and generation of control instructions through the coordinated operation of a multidimensional parameter coordination unit, a dynamic gradient optimization unit, and an abnormal trajectory identification unit. The multidimensional parameter coordination unit comprises multiple state decision nodes, each of which connects to eigenvectors in the compensation field data and path coupling feature data through configuration parameters. For example, for a state decision node related to temperature control, the connected eigenvectors might include the temperature parameter from the reaction condition characteristics, the temperature-sensitive chemical bond energy parameters (second quantitative labels) from the reactant physical properties, and the temperature-affected active site parameters (local features in the third mapping sequence) from the catalyst distribution characteristics. Configuration parameters define the node's response thresholds and weights for different eigenvectors. For example, the configuration parameters for a temperature node might be set such that a temperature warning instruction is triggered when the temperature eigenvector value exceeds a preset high-temperature threshold and the corresponding chemical bond energy eigenvector indicates a significant increase in reaction activity.

[0087] The dynamic gradient optimization unit iteratively optimizes the configuration parameters using a dynamic gradient adjustment algorithm to minimize the error between the reaction state command and the actual reaction trajectory. This process is based on the gradient descent principle of the loss function L, which can be expressed as:

[0088]

[0089] in, is the sample size, For the The reaction state instruction prediction vector of samples, is the true state vector corresponding to the actual reaction trajectory, Represents the L2 norm. The dynamic gradient optimization unit calculates the loss function to optimize the configuration parameters. Gradient , and update the parameters in the opposite direction of the gradient:

[0090]

[0091] Where, is the learning rate, is the number of iterations. Through repeated iterations, the configuration parameters are gradually optimized, bringing the predicted reaction state instructions closer to the actual reaction trajectory. For example, when simulating an esterification reaction, if the initial configuration parameters result in a deviation between the predicted reaction rate and the actual rate, the dynamic gradient optimization unit will automatically adjust the weights of the temperature and concentration nodes until the error converges.

[0092] The abnormal trajectory identification unit locates high-risk reactions based on compensation field data and path coupling feature data. This unit first constructs a high-risk feature space, mapping multidimensional parameters (such as energy mutation rate, abnormal fluctuation amplitude of material concentration, and reaction rate deviation threshold) into the feature space. It then uses a density clustering algorithm (such as DBSCAN) to identify outliers, which correspond to high-risk reaction nodes. In specific implementation, the system defines the energy mutation rate as:

[0093]

[0094] in, is the total energy of the reaction system at the current moment, for The energy value before time steps, is the time step. Exceeding the preset threshold At the same time, combined with the material concentration characteristic vector The concentration change rate of key species (such as flammable and explosive intermediates) in ,If both exceed the safety range at the same time, it is determined as a high-risk ,reaction trajectory, and an emergency stop command or a safety ,protection start command is generated.

[0095] The state decision nodes of the multi-dimensional parameter coordination unit adopt a layered architecture design. Bottom-level nodes are responsible for single-dimensional characteristic responses (such as temperature and pressure nodes). Higher-level nodes combine the outputs of these nodes through logic gates (AND, OR, and NOT) to determine complex reaction states. For example, the high-level node "Overtemperature and Overpressure Warning" is formed by combining the temperature node (exceeding the high temperature threshold) and the pressure node (exceeding the high pressure threshold) through AND logic. Only when both are triggered simultaneously will an overtemperature and overpressure warning instruction be generated. This layered architecture improves the system's ability to analyze complex reaction states and avoids erroneous instructions caused by misjudgment of a single parameter.

[0096] The dynamic gradient optimization unit introduces momentum factors during the iteration process To accelerate convergence, the parameter update formula is adjusted to:

[0097]

[0098]

[0099] in, This is a momentum accumulation term that retains historical gradient information, reducing oscillations during parameter updates and improving optimization efficiency. This mechanism is particularly useful in scenarios where reaction data is noisy or multimodal. For example, in catalytic reactions, random deactivation of catalyst active sites causes reaction trajectories to fluctuate. The momentum factor can help the system converge more stably to the optimal configuration parameters.

[0100] The abnormal trajectory recognition unit also integrates a time series prediction model, using historical data to train a long short-term memory (LSTM) network to predict future reaction trajectories. When the predicted trajectory enters the warning zone of the high-risk feature space, the system generates preventative control instructions in advance, even if the current data has not yet triggered the abnormal threshold. For example, in a polymerization reaction, if the LSTM model predicts that energy accumulation will exceed the safety threshold within the next 30 minutes, the abnormal trajectory recognition unit will trigger a pre-start instruction for the cooling system to prevent the reaction from running out of control.

[0101] The three units of the state trigger layer achieve real-time data interaction through the data bus. The preliminary instructions generated by the multi-dimensional parameter coordination unit are passed to the dynamic gradient optimization unit for error assessment and parameter adjustment. The optimized instructions are then fed back to the multi-dimensional parameter coordination unit to update the node status. The abnormal trajectory identification unit continuously monitors the compensation field data and path coupling feature data. Once a high-risk signal is detected, it immediately sends an interrupt request to the multi-dimensional parameter coordination unit and generates emergency instructions first. This interactive mechanism ensures that the system can operate efficiently in both conventional simulation and emergency response scenarios, which not only guarantees the accuracy of the reaction simulation, but also improves the safety and robustness of the system.

[0102] In its implementation, the configuration parameters of the state decision nodes are stored in a dynamically updateable parameter table, allowing experimenters to manually adjust initial parameter values ​​based on different reaction types. For example, for highly exothermic reactions, experimenters can pre-increase the weight coefficient and warning threshold of the temperature node to make the system more sensitive to temperature changes. The dynamic gradient optimization unit runs continuously in the background, leveraging historical reaction data to continuously optimize the parameter table, developing adaptive learning capabilities and gradually reducing reliance on manual intervention.

[0103] In summary, the state trigger layer builds a control module that integrates intelligent decision-making, adaptive learning, and safety protection through the collaborative analysis of multi-dimensional parameters, dynamic optimization of configuration parameters, and real-time identification of high-risk trajectories. This module not only generates accurate state instructions based on real-time reaction data but also improves predictive capabilities through continuous learning, providing a reliable control center for chemical reaction experiment simulations.

[0104] Example 5:

[0105] In the AI-based chemical reaction experiment simulation system, the reaction process monitoring and safety warning module achieves comprehensive monitoring and intelligent early warning of chemical reaction processes through four core links: multi-dimensional data acquisition, real-time feature extraction, abnormal pattern recognition, and graded early warning response. The specific implementation of this module is explained in detail using the typical case of an esterification reaction experiment.

[0106] In the multi-dimensional data collection phase, the system deploys various types of sensors to obtain comprehensive information about the reaction process. Temperature sensors are arranged in a distributed manner, with a high-precision thermocouple installed at the top, middle, and bottom of the reaction vessel to monitor temperature changes at different locations in the reaction system in real time. A pressure sensor is installed at the top of the reaction vessel to measure pressure fluctuations during the reaction. The concentration sensor uses spectral analysis technology to penetrate the reaction solution with light of a specific wavelength and calculate the concentrations of reactants and products based on changes in absorbance. In addition, the system is equipped with a pH sensor to monitor changes in the acidity and alkalinity of the reaction solution, as well as a stirring rate sensor to obtain the working status of the stirring device in real time. These sensors collect data every 0.1 seconds, forming a multi-dimensional real-time data stream that provides basic data support for subsequent reaction monitoring.

[0107] The real-time feature extraction process processes and analyzes the collected raw data, extracting key features that reflect the reaction status. The system first processes the temperature data, calculating the temperature change rate—the temperature difference between two adjacent time points divided by the time interval. It also analyzes the uniformity of the temperature distribution and assesses the heat conduction of the reaction system by comparing the data differences from temperature sensors at different locations. For pressure data, the system extracts the frequency and amplitude characteristics of pressure fluctuations to identify pressure change trends. For concentration data processing, the system calculates the consumption rate of reactants and the production rate of products, as well as the proportional relationship between the two. Furthermore, the system performs real-time analysis of changes in pH and stirring rate, extracting pH trend changes and stirring rate stability characteristics. These characteristic data can intuitively reflect the progress and status of the reaction, providing a basis for identifying abnormal patterns.

[0108] The abnormal pattern recognition phase uses machine learning algorithms based on extracted feature data to identify abnormal patterns in the reaction process. The system pre-trains multiple anomaly detection models, including a threshold model based on statistical distribution, an outlier detection model based on cluster analysis, and a time series prediction model based on deep learning. In the esterification reaction experiment, the system first uses the threshold model to monitor key parameters such as temperature and pressure. For example, when the rate of temperature change exceeds a preset safety threshold, the system initially determines that an anomaly may exist. To improve the accuracy of this judgment, the system further uses an outlier detection model to analyze the multi-dimensional feature data. This model compares the current feature data with the normal reaction pattern and calculates the degree of deviation of the data point from the normal pattern. If the deviation exceeds a set threshold, an anomaly is considered. The system also uses a time series prediction model to predict the reaction status. This model uses historical data to predict the changes in reaction parameters over a period of time. If the error between the actual measured value and the predicted value exceeds a certain range, an anomaly warning is issued.

[0109] The graded warning response process generates warnings of varying levels and implements corresponding response measures based on the severity and development trend of the anomaly. The system categorizes warning levels into four levels: Level 1 for minor anomalies, Level 2 for moderate anomalies, Level 3 for severe anomalies, and Level 4 for emergencies. During an esterification reaction experiment, if the system detects a slight increase in the temperature change rate, but still within an acceptable range, a Level 1 warning will be issued. The system will automatically record the anomaly and alert the experimenter to the reaction status. If the temperature continues to rise and the pressure begins to fluctuate abnormally, a Level 2 warning will be issued. The system will then automatically adjust reaction conditions, such as reducing heating power and increasing stirring rate, and send a detailed anomaly report to the experimenter. If the temperature and pressure rise sharply, exceeding safety thresholds, the system will issue a Level 3 warning. The system will immediately activate the emergency cooling system and send an alert to the experimenter and laboratory management. If the anomaly worsens and could potentially lead to a dangerous situation, the system will issue a Level 4 warning. The system will then automatically cut off the power and reactant supply and activate safety devices, such as explosion-proof valves and sprinkler systems, to ensure the safety of experimenters and equipment.

[0110] The system also possesses self-learning and optimization capabilities throughout the entire reaction process monitoring and safety warning process. By analyzing and summarizing historical anomaly data, the system continuously optimizes anomaly detection models and warning thresholds, improving the accuracy and reliability of warnings. Furthermore, the system adjusts warning strategies and response measures based on feedback from experimenters, making it more adaptable to different experimental needs and scenarios. For example, after multiple esterification reaction experiments, the system adjusts the temperature and pressure warning thresholds based on actual conditions to better suit the characteristics of the reaction.

[0111] The reaction process monitoring and safety warning module achieves comprehensive monitoring and intelligent early warning of chemical reaction processes through the coordinated work of four core components: multi-dimensional data acquisition, real-time feature extraction, anomaly pattern recognition, and graded early warning response. This module promptly detects anomalies during the reaction process and takes appropriate measures based on their severity, effectively improving the safety and reliability of chemical reaction experiments. Through continuous self-learning and optimization, the system's early warning capabilities and adaptability will continue to improve, providing more reliable protection for chemical reaction experiments.

[0112] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0113] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based chemical reaction experiment simulation system, characterized in that: include: A reaction mechanism model for performing multi-dimensional parameter integration processing on initial experimental data, wherein the initial experimental data includes a first eigenvector corresponding to the reactant physical property characteristic data, a second correlation matrix corresponding to the reaction condition characteristic data, and a third mapping sequence corresponding to the catalyst distribution characteristic data. The reactant physical property characteristic data includes a first distribution label generated by molecular structure topology parameters and a second quantitative label generated by chemical bond energy parameters. A dynamic simulation model is used to perform dynamic disturbance suppression processing on the real-time reaction data stream and input it into the path generation layer to simulate the reaction trajectory, and generate reaction state instructions according to the output results of the path generation layer; The path generation layer includes a data preprocessing module and a dynamic path generation module. The data preprocessing module is used to split the original reaction data stream into energy intervals and filter invalid features. The dynamic path generation module is obtained through collaborative training based on multiple types of historical reaction data and real-time environmental parameters. The dynamic path generation module includes a path parsing layer, a path correction layer, and a state triggering layer connected in sequence. The path parsing layer is used to perform energy level correlation processing on different eigenvectors in the original reaction data stream to generate path coupling feature data; the path correction layer is used to model the dynamic matching relationship between the path coupling feature data corresponding to each eigenvector to generate compensation field data; the state triggering layer is used to perform multi-dimensional parameter fusion based on the compensation field data and the path coupling feature data to generate reaction state instructions.

2. The AI-based chemical reaction experiment simulation system according to claim 1, characterized in that: The modeling of the dynamic matching relationship between the path coupling characteristic data corresponding to each characteristic vector to generate compensation field data includes: A path segmentation algorithm is used to identify the key reaction nodes in the path coupling characteristic data, and a correction compensation sequence corresponding to each characteristic vector is determined based on the environmental parameter type corresponding to each key reaction node; The deviation coefficient between the reaction nodes of the same parameter in the correction compensation sequence corresponding to any two eigenvectors is calculated, and the compensation field data between the any two eigenvectors is generated based on the deviation coefficient.

3. The AI-based chemical reaction experiment simulation system according to claim 2, characterized in that: The calculation of the deviation coefficient between the reaction nodes with the same parameters in the correction compensation sequence corresponding to any two eigenvectors includes: When the number of reaction nodes in the correction compensation sequence corresponding to any two eigenvectors is inconsistent, virtual reaction point interpolation is performed based on the environmental parameters corresponding to the terminal reaction node among the reaction nodes with fewer reaction nodes, and the deviation coefficient between the reaction nodes with the same parameters is calculated based on the interpolated data.

4. The AI-based chemical reaction experiment simulation system according to claim 1, characterized in that: The data preprocessing module is specifically used for: Performing isogradient division on the first eigenvector, the second correlation matrix, and the third mapping sequence according to a preset energy interval to generate standardized reaction data, standardized condition data, and standardized catalytic data; A dynamic parameter matching method is used to align the standardized reaction data and the standardized condition data in real time, and a fixed energy window mechanism is used to perform steady-state optimization on the standardized catalytic data, outputting a first calibration vector, a second calibration matrix, and a third calibration sequence; wherein the first calibration vector includes a calibrated first distribution label and a calibrated second quantization label.

5. The AI-based chemical reaction experiment simulation system according to claim 4, characterized in that: The data preprocessing module is also used for: Calculating a path deviation coefficient between the calibrated first distribution label and the calibrated second quantization label within a historical reaction period; Predicting an expected distribution value of the calibrated second quantized label in the real-time response phase according to the path deviation coefficient and the parameter value of the calibrated first distribution label in the real-time response phase; Target correction compensation data is generated based on the calibrated second quantization label and its expected distribution value, and a feature vector corresponding to the target correction compensation data is used as a first calibration vector.

6. The AI-based chemical reaction experiment simulation system according to claim 1, characterized in that: The path correction layer specifically includes: a path tracing unit, configured to perform reaction path tracing on each characteristic vector in the path coupling characteristic data, so as to extract a corresponding reaction propagation chain from each characteristic vector; The parameter matching unit is used to perform energy level superposition on the reaction propagation chain extracted by each eigenvector and the corresponding path coupling characteristic data to generate compensation field data.

7. The AI-based chemical reaction experiment simulation system according to claim 6, characterized in that: The path correction layer also includes: The noise elimination unit is used to perform phase lag effect elimination processing on the compensation field data.

8. The AI-based chemical reaction experiment simulation system according to claim 1, characterized in that: The state trigger layer specifically includes: A multi-dimensional parameter coordination unit, comprising a plurality of state decision nodes, each state decision node being connected to each characteristic vector in the compensation field data and the path coupling characteristic data through configuration parameters; a dynamic gradient optimization unit, configured to iteratively optimize the configuration parameters using a dynamic gradient adjustment algorithm to minimize the error between the reaction state command and the actual reaction trajectory; The abnormal trajectory identification unit is used to locate high-risk reactions based on the compensation field data and the path coupling characteristic data, and generate reaction state instructions.

9. An AI-based chemical reaction experiment simulation method, applied to the AI-based chemical reaction experiment simulation system according to any one of claims 1 to 8, characterized in that: The following steps are involved: Performing multi-dimensional parameter integration processing on initial experimental data using a reaction mechanism model, wherein the initial experimental data includes a first eigenvector corresponding to reactant physical property characteristic data, a second correlation matrix corresponding to reaction condition characteristic data, and a third mapping sequence corresponding to catalyst distribution characteristic data, wherein the reactant physical property characteristic data includes a first distribution label generated by molecular structure topology parameters and a second quantitative label generated by chemical bond energy parameters; The real-time reaction data stream is subjected to dynamic disturbance suppression processing through a dynamic simulation model, and the processed data stream is input into the path generation layer for reaction trajectory simulation; The data preprocessing module in the path generation layer is used to split the original reaction data stream into energy intervals and filter invalid features; The reaction trajectory simulation is performed using a dynamic path generation module obtained by collaborative training of multiple types of historical reaction data and real-time environmental parameters. The dynamic path generation module includes a path parsing layer, a path correction layer, and a state triggering layer connected in sequence. According to the output results of the path generation layer, the dynamic simulation model generates reaction state instructions; The path parsing layer is used to perform energy level correlation processing on different eigenvectors in the original reaction data stream to generate path coupling feature data; the path correction layer is used to model the dynamic matching relationship between the path coupling feature data corresponding to each eigenvector to generate compensation field data; the state triggering layer is used to perform multi-dimensional parameter fusion based on the compensation field data and the path coupling feature data to generate reaction state instructions.

Images