AI-driven chemical experiment simulation and result prediction system

Abstract

The present invention relates to the field of chemical experiment technology and discloses an AI-driven chemical experiment simulation and result prediction system. The system collects multimodal data through an experimental data acquisition module, extracts spectral features through a spectral feature analysis module, constructs a coupled kinetic model through a reaction kinetic modeling module, predicts product distribution probabilities and optimizes reaction conditions through a dynamic result prediction module, and adjusts reaction parameters in real time through an experimental parameter feedback module. In addition, the system also includes a reaction safety assessment model for identifying risks and providing safety protection solutions. The system realizes comprehensive simulation and accurate result prediction of chemical experiments, can effectively optimize reaction conditions, and improve reaction efficiency and safety. It has broad application prospects in fields such as chemical experimental research and chemical production.

Description

Technical Field

[0001] The present invention relates to the field of chemical experiment technology, and specifically to an AI-driven chemical experiment simulation and result prediction system. Background Art

[0002] In the field of chemical experiments, traditional experimental methods face many challenges and are unable to meet the growing needs of modern chemical research and industrial production. From the perspective of experimental data acquisition, in the past, data collection mainly relied on a single type of sensor, and the data dimension was limited. For example, when studying solution reactions, only the basic parameter of reactant concentration can be measured, and it is difficult to fully obtain key information such as solution temperature distribution and gas diffusion dynamics. This makes it impossible for researchers to gain an in-depth understanding of the microscopic changes and complex interaction mechanisms in the reaction process. Although the application of multispectral imaging technology and chemical sensor arrays has gradually emerged, there are still deficiencies in the systematic and collaborative nature of data acquisition, and it is impossible to efficiently integrate multimodal data, resulting in a large amount of valuable information being missed.

[0003] The prediction and analysis of experimental results also face bottlenecks. Relying on experience and simple mathematical models to predict results suffers from poor accuracy and reliability. Taking complex organic synthesis reactions as an example, due to the complex reaction mechanisms and numerous influencing factors, traditional prediction methods often fail to account for the significant impact of small changes in reaction conditions on product distribution, resulting in significant deviations between predicted results and actual experimental results. When faced with new chemical reaction systems, traditional methods are even more difficult to quickly generate effective prediction and optimization solutions, seriously hindering the research and development of new chemical products.

[0004] The control accuracy and real-time adjustment capabilities of experimental conditions also restrict the development of chemical experiments. Traditional experimental devices usually adopt fixed control strategies and cannot dynamically adjust reaction conditions according to real-time changes in the reaction process. For example, in terms of temperature control, it is difficult to achieve precise regulation of the temperature gradient of the reaction system. Once temperature fluctuations occur during the reaction, it is impossible to respond in time, resulting in an unstable reaction environment, affecting reaction efficiency and product quality. In terms of stirring rate control, there is also a lack of means to make intelligent adjustments according to the reaction process, which cannot fully promote the mixing and mass transfer of reactants, limiting the progress of the reaction.

[0005] Furthermore, the increasing adoption of green chemistry and sustainable development concepts has led to higher demands for the safety, environmental friendliness, and resource efficiency of chemical experiments. However, traditional experimental methods are relatively weak in reaction safety assessments, lacking real-time monitoring and effective early warning mechanisms for runaway reaction risks, which can easily lead to safety accidents. Furthermore, the inability to precisely optimize reaction conditions often results in waste of reactants, increasing production costs while also negatively impacting the environment. Summary of the Invention

[0006] The purpose of the present invention is to provide an AI-driven chemical experiment simulation and result prediction system to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an AI-driven chemical experiment simulation and result prediction system, the system comprising:

[0008] Experimental data acquisition module: used to collect multimodal data of the experimental process through a chemical sensor array and a multispectral imaging device. The multimodal data includes reactant concentration time series data, solution temperature distribution images, gas diffusion dynamic video streams, and reaction vessel pressure waveforms;

[0009] Spectral feature analysis module: uses wavelet transform combined with convolutional neural network to extract spectral features of the multimodal data, generating multi-dimensional analysis results including material absorption peak position characteristics, reaction rate correlation characteristics and phase change process spatiotemporal characteristics;

[0010] Reaction kinetics modeling module: Based on the multi-dimensional analytical results, a coupled kinetic model of the reactant concentration field and the temperature field is constructed using partial differential equations, and the gradient change of the reaction path is solved by the finite element method;

[0011] Dynamic result prediction module: Based on the coupled kinetic model, a time-series convolutional network is used to predict the product distribution probability under different experimental parameters and generate a reaction condition optimization plan;

[0012] Experimental parameter feedback module: According to the reaction condition optimization plan, the temperature gradient and stirring rate parameters of the reaction system are adjusted in real time through an adaptive control algorithm.

[0013] Preferably, the experimental data acquisition module includes:

[0014] An electrochemical sensor array was used to measure the time series data of ion mobility at the reaction interface, and an infrared thermal imager was used to capture the surface temperature distribution of the reaction vessel.

[0015] A high-speed camera was used to record the turbulent characteristic video stream of the gas evolution process, and the oscillation waveform data of the pressure sensor in the reactor was collected synchronously.

[0016] Preferably, the spectral feature analysis module further includes:

[0017] A frequency domain decomposition algorithm is used to separate the spectral responses of different bands of the multispectral imaging data, and a weighted fusion feature channel is performed in combination with an attention mechanism;

[0018] The vortex structure characteristics of the gas diffusion dynamic video flow are extracted by a three-dimensional convolution kernel, and a correlation matrix between turbulence intensity and reaction rate is constructed.

[0019] Preferably, the reaction kinetics modeling module further includes:

[0020] Establish a non-steady-state mass transfer equation to describe the concentration gradient of reactants, and input the temperature field data as the boundary condition;

[0021] A multi-grid iterative algorithm is used to accelerate the partial differential equation solution process, and a deep neural network proxy model is constructed based on residual connections.

[0022] Preferably, the dynamic result prediction module further includes:

[0023] A bidirectional gated recurrent unit network is constructed to model the long-term dependencies of the reaction process, extract multi-scale representations of temporal features, generate a perturbation set of reaction parameters through Monte Carlo sampling, and calculate the probability density function of the product distribution.

[0024] Preferably, the dynamic result prediction module further includes:

[0025] Generative adversarial networks are used to construct a virtual experimental environment to generate simulation data of reaction processes under extreme conditions.

[0026] The feature distribution differences between virtual data and actual observed data are aligned through contrastive learning algorithm.

[0027] Preferably, the experimental parameter feedback module further comprises: designing a fuzzy PID controller to regulate the power output of the constant temperature device, and dynamically adjusting the control parameters based on the gradient characteristics of the temperature distribution diagram.

[0028] Preferably, the experimental parameter feedback module further includes:

[0029] A reinforcement learning framework is used to train the parameter adjustment strategy network, defining the state space as the multimodal feature vector of the current reaction system;

[0030] The reward function constrains the balance between energy consumption and reaction efficiency to generate the optimal control instruction sequence.

[0031] Preferably, the system further comprises:

[0032] A reaction safety assessment model is constructed, and based on the gradient field data of the coupled dynamics model, a support vector machine classifier is used to identify the reaction runaway risk threshold; when risk features are detected, an emergency parameter adjustment protocol is triggered and a safety protection recommendation plan is generated.

[0033] Preferably, the reaction safety assessment model further includes: associating historical accident case data through a knowledge graph to generate a safety protection strategy reasoning path for the current reaction system.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The AI-driven chemical experiment simulation and result prediction system provided by the present invention has significant beneficial effects in many aspects. In terms of experimental data acquisition, through the chemical sensor array and multispectral imaging device, multimodal data such as reactant concentration time series data, solution temperature distribution images, gas diffusion dynamic video streams, and reaction vessel pressure waveforms can be obtained. This comprehensive data acquisition method greatly enriches researchers' cognitive dimensions of the reaction process. Taking catalytic reaction experiments as an example, in the past, only the approximate changes in reactant concentration over time could be known. Now, with the help of this system, not only can the details of concentration changes be accurately tracked, but the real-time distribution of temperature in the reaction system can also be observed simultaneously, and the dynamic process of gas diffusion and pressure fluctuations can be understood. This helps to deeply explore the microscopic mechanisms of the reaction process and provides a richer and more accurate data foundation for revealing the essential laws of chemical reactions.

[0036] The spectral feature analysis module uses wavelet transforms combined with convolutional neural networks to extract spectral features and generate multi-dimensional analysis results, enabling more in-depth and accurate analysis of experimental data. The acquisition of information such as the positional characteristics of absorption peaks, reaction rate correlations, and the spatiotemporal characteristics of phase transitions can help researchers better understand the changes and progression of substances in chemical reactions. When analyzing the synthesis reaction of a new material, these characteristics can accurately identify key stages of the reaction, predict the direction of the reaction in advance, and provide a strong basis for timely adjustments to experimental plans.

[0037] The reaction kinetics modeling module constructs a coupled kinetic model based on multi-dimensional analytical results and uses the finite element method to solve for gradient changes in the reaction path. This modeling approach can more accurately describe the interaction between the reactant concentration field and the temperature field during the reaction process. In large-scale reaction simulations in chemical production, this model can help engineers optimize the design of reaction equipment, rationally arrange the feed location and temperature distribution of reactants, improve reaction efficiency, and reduce production costs. Furthermore, based on the calculation results of this model, energy changes during the reaction process can be more accurately predicted, providing a scientific basis for thermal management of the reaction.

[0038] The dynamic result prediction module uses a temporal convolutional network to predict product distribution probabilities under different experimental parameters and generate optimized reaction condition scenarios. This provides efficient decision support for chemical experiments and chemical production. In drug synthesis experiments, researchers can use the predicted results to quickly find the optimal combination of parameters such as reaction temperature, pressure, and reactant ratio, reducing the number of trial and error experiments and significantly shortening the R&D cycle. In industrial production, optimizing reaction conditions can improve product quality and yield, enhancing a company's market competitiveness.

[0039] The experimental parameter feedback module adjusts the reaction system's temperature gradient and stirring rate parameters in real time based on the reaction condition optimization plan, ensuring that the reaction always proceeds under optimal conditions. This real-time feedback and adjustment mechanism effectively improves reaction stability and controllability. In fine chemical production, even minor changes in reaction conditions can affect product quality. This module can promptly adjust parameters based on reaction conditions, ensuring consistent product quality and reducing scrap rates.

[0040] Furthermore, the system's reaction safety assessment model identifies runaway reaction risk thresholds based on gradient field data from coupled kinetic models. When risk signatures are detected, it triggers emergency parameter adjustment protocols and generates safety protection recommendations. By linking historical accident case data with a knowledge graph, it generates a safety protection strategy inference path tailored to the current reaction system. This significantly improves the safety of chemical experiments and production processes, effectively avoiding safety incidents caused by runaway reactions and safeguarding both personnel and enterprise assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a working principle diagram of the AI-driven chemical experiment simulation and result prediction system of the present invention;

[0042] Figure 2 A detailed flowchart for the experimental data acquisition module;

[0043] Figure 3 A diagram showing the working principle for further expansion of the dynamic result prediction module;

[0044] Figure 4 This is a working principle diagram for further expansion of the experimental parameter feedback module. DETAILED DESCRIPTION

[0045] 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.

[0046] See also Figures 1-4 The present invention provides an AI-driven chemical experiment simulation and result prediction system, the specific implementation process of which is as follows:

[0047] The experimental data acquisition module collects multimodal data from the experimental process. This module utilizes a chemical sensor array and a multispectral imaging device to collect multimodal data, including reactant concentration time series data, solution temperature distribution images, dynamic gas diffusion video streams, and reaction vessel pressure waveforms. For example, during a specific chemical reaction experiment, the chemical sensor array monitors the changes in reactant concentration over time in real time, generating reactant concentration time series data. The multispectral imaging device then captures the reaction solution and obtains solution temperature distribution images, visually demonstrating the solution temperature distribution during the reaction. Simultaneously, related equipment is used to record the dynamic gas diffusion video stream and reaction vessel pressure waveforms, thereby comprehensively collecting various data from the experimental process.

[0048] The spectral feature analysis module processes the collected multimodal data. It uses wavelet transforms combined with convolutional neural networks to extract spectral features, generating multidimensional analysis results that include the positional characteristics of the substance's absorption peaks, reaction rate correlations, and the spatiotemporal characteristics of the phase transition process. This process facilitates in-depth analysis of the substance's changes during the reaction, providing an important basis for subsequent reaction kinetic modeling.

[0049] The reaction kinetics modeling module operates based on multi-dimensional analytical results. It uses partial differential equations to construct a coupled kinetic model of the reactant concentration and temperature fields, describing the relationship between reactant concentration and temperature during the reaction. Finite element methods are then used to solve for gradient changes in the reaction path, providing a more precise understanding of the reaction's dynamics.

[0050] The dynamic result prediction module then performs predictions based on the coupled kinetic model. It uses a time-series convolutional network to predict the product distribution probability under different experimental parameters, providing a basis for predicting experimental results. The module also generates reaction condition optimization plans to guide experimenters in adjusting experimental conditions to achieve more optimal results.

[0051] The experimental parameter feedback module adjusts the reaction system in real time based on the reaction condition optimization plan. Through an adaptive control algorithm, the temperature gradient and stirring rate parameters of the reaction system are adjusted in real time to ensure that the reaction proceeds under optimized conditions, improving reaction efficiency and product quality.

[0052] The technical solution of the present invention is further described in detail below with reference to specific embodiments.

[0053] Example 1:

[0054] In this example, the implementation of the experimental data acquisition module is further refined. Specifically, the module uses an electrochemical sensor array to measure time-series data on ion mobility at the reaction interface and an infrared thermal imager to capture the surface temperature distribution of the reaction vessel. Simultaneously, a high-speed camera records a video stream of the turbulent flow characteristics of the gas evolution process and simultaneously collects oscillation waveform data from the pressure sensor within the reactor.

[0055] When measuring the time series data of ion mobility at the reaction interface using an electrochemical sensor array, the electrochemical sensor array is composed of multiple different types of electrochemical sensors that can accurately measure the migration of ions at the reaction interface. Its working principle is based on the ion conduction characteristics of electrochemistry. When a reaction occurs, ions migrate at the reaction interface. The electrochemical sensor can detect the changes in the electrical signal generated by ion migration in real time and convert it into ion mobility data, recording it in chronological order to form the time series data of ion mobility at the reaction interface. By analyzing this data, we can understand the dynamic changes of ions during the reaction process, providing important information for studying the reaction mechanism.

[0056] When using an infrared thermal imager to capture the surface temperature distribution of a reaction vessel, the camera uses the infrared radiation properties of an object to measure temperature. During the reaction process, the surface temperature of the reaction vessel changes due to the thermal effects of the chemical reaction. The infrared thermal imager receives infrared radiation emitted by the reaction vessel surface, converts it into an electrical signal, and then, through a series of signal processing and algorithm conversion, generates a surface temperature distribution map of the reaction vessel. This map uses different colors to intuitively display the temperature differences in different parts of the reaction vessel surface, helping experimenters clearly observe the temperature distribution during the reaction and analyze the impact of temperature on the reaction.

[0057] When using a high-speed camera to record a video stream of turbulent flow characteristics during gas evolution, the camera captures the gas evolution area at a high frame rate. Gas evolution generates complex turbulent flows, and the high-speed camera captures the transient state of these turbulent flows and records them as a video stream. Analysis of the video stream allows the extraction of various turbulent characteristics, such as velocity and vortex structure. These characteristics are important for studying gas diffusion and reaction kinetics. To simultaneously collect oscillation waveform data from a pressure sensor within the reactor, the pressure sensor is installed at a suitable location within the reactor. When the pressure within the reactor changes due to gas production or consumption during the reaction, the pressure sensor senses the pressure fluctuations and converts them into electrical signals. These electrical signals are processed to generate oscillation waveform data, reflecting the temporal changes in pressure during the reaction. Analysis of the pressure oscillation waveform data can provide information on the progress of the reaction and the rate of gas production or consumption.

[0058] Example 2:

[0059] This example details the implementation of the spectral feature analysis module. This module not only extracts spectral features from multimodal data using wavelet transforms combined with convolutional neural networks, but also uses a frequency-domain decomposition algorithm to separate the spectral responses of different bands in multispectral imaging data, and then integrates an attention mechanism to weightedly fuse feature channels. Furthermore, a three-dimensional convolution kernel is used to extract vortex structure features from dynamic gas diffusion video streams, constructing a correlation matrix between turbulence intensity and reaction rate.

[0060] When the frequency domain decomposition algorithm is used to separate the spectral responses of different bands of multispectral imaging data, the multispectral imaging data contains spectral information of multiple bands. The frequency domain decomposition algorithm is based on the Fourier transform principle to convert the multispectral imaging data from the time domain to the frequency domain. In the frequency domain, the spectral responses of different bands have different frequency characteristics. By setting appropriate frequency filters, the spectral responses of different bands can be separated. Assuming that the multispectral imaging data is , after Fourier transformation Get its frequency domain representation , and then according to the frequency range corresponding to different bands , etc., using a bandpass filter , equal Filtering, that is , etc., and then undergo inverse Fourier transform , The spectral response data of different bands after separation are obtained , Etc. In this way, the separation of spectral responses in different bands is achieved, which provides a basis for subsequent feature extraction and analysis.

[0061] When combined with the attention mechanism to weightedly fuse feature channels, the attention mechanism is a method that allows the model to pay more attention to important features. In spectral feature analysis, the information contained in the spectral response data of different bands has different importance for reaction analysis. First, feature extraction is performed on the separated spectral response data of each band to obtain the feature vector , ,..., Then, the weight of each feature vector is calculated through an attention network , ,..., , the attention network is usually composed of a multi-layer neural network, whose input is the concatenation of all feature vectors , the output is the weight vector Finally, the weights are weighted and fused with the feature vector to obtain the fused feature vector In this way, important features are given higher weights, making the fused features more able to reflect the key information of the response.

[0062] When extracting the vortex structure features of the gas diffusion dynamic video stream through the three-dimensional convolution kernel, the gas diffusion dynamic video stream is a three-dimensional data containing information in two dimensions: time and space. The three-dimensional convolution kernel can perform convolution operations on the video stream data simultaneously in time and space. Let the three-dimensional convolution kernel be , whose size is ,in represents the convolution kernel size in the time dimension, and Indicates the size of the convolution kernel in the spatial dimension. For video stream data , after the three-dimensional convolution operation, the feature map is obtained By designing a suitable three-dimensional convolution kernel, the vortex structure features in the gas diffusion dynamic video flow can be effectively extracted, and these features reflect the turbulent characteristics of the gas diffusion process.

[0063] When constructing the correlation matrix between turbulence intensity and reaction rate, the turbulence intensity needs to be calculated first. Based on the extracted vortex structure characteristics, the turbulence intensity can be calculated through a certain algorithm. Reaction rate It can be obtained by measuring the rate of change of reactant concentration over time, that is, ,in is the reactant concentration, is time. Then, the turbulence intensity at different times is , ,..., The corresponding reaction rate , ,..., Make associations and build an association matrix , the matrix elements Indicates the The turbulence intensity at the moment is The correlation between the reaction rates at different moments, such as By analyzing this correlation matrix, we can gain a deeper understanding of the intrinsic relationship between turbulence intensity and reaction rate, providing strong support for reaction kinetics research.

[0064] Example 3:

[0065] This example provides an in-depth explanation of the reaction kinetics modeling module. Based on multidimensional analytical results, the module uses partial differential equations to construct a coupled kinetic model of the reactant concentration and temperature fields. Furthermore, the module uses the finite element method to solve the gradient changes in the reaction path. Furthermore, the module establishes an unsteady-state mass transfer equation to describe the gradient changes in reactant concentrations, using temperature field data as boundary conditions. Furthermore, a multigrid iterative algorithm is used to accelerate the partial differential equation solution process, and a deep neural network proxy model is constructed based on residual connections.

[0066] When establishing a non-steady-state mass transfer equation to describe the change in reactant concentration gradient, the non-steady-state mass transfer equation is established based on the principle of conservation of matter. Assume that the reactant concentration is , time is , the spatial coordinates are , , , the diffusion coefficient is The general form of the unsteady mass transfer equation is . This equation describes the diffusion of reactants in space over time, where the left side of the equation represents the rate of change of reactant concentration over time, and the right side of the equation represents the concentration change due to diffusion. In actual reaction systems, the diffusion of reactants is affected by many factors, and this equation can be used to quantitatively analyze these influences. When temperature field data is input as a boundary condition, because temperature affects parameters such as the diffusion coefficient of the reactants, the temperature field data needs to be incorporated into the unsteady-state mass transfer equation. Assume that the temperature field data is , the temperature and diffusion coefficient can be related through certain functional relationships, such as ,in is the reference diffusion coefficient, is the diffusion activation energy, is the gas constant, In this way, the temperature field data participates in the unsteady mass transfer equation by affecting the diffusion coefficient, and more accurately describes the concentration changes during the reaction process.

[0067] The multigrid iterative algorithm is an efficient numerical solution method for accelerating the solution of partial differential equations. For partial differential equations, the solution process usually requires discretizing the entire computational domain, forming a large system of linear equations. The multigrid iterative algorithm quickly approximates the exact solution of the equation by iteratively solving on grids of different resolutions. First, a preliminary iteration is performed on the finest grid to obtain an approximate solution. Due to the presence of high-frequency errors on the fine grid, the error on the fine grid is transferred to the coarse grid through a restriction operator, and the error is corrected on the coarse grid. Then, the corrected result on the coarse grid is transferred back to the fine grid through an extension operator, and the iteration continues. In this way, by repeatedly iterating between different grids, the solution process can be effectively accelerated and the calculation time can be reduced.

[0068] When building a deep neural network proxy model based on residual connection, residual connection is a commonly used structure in deep neural networks. Assume that the input of the deep neural network is , the output is , the input of a layer in the network is , the output is The output of a traditional neural network layer is , while in the residual connection, ,in is the nonlinear transformation function of this layer. This residual connection approach effectively solves the vanishing gradient problem during deep neural network training, allowing the network to more easily learn complex mapping relationships. When constructing a deep neural network proxy model, the inputs of the reaction kinetics model (such as reactant concentrations and temperature) are used as the input of the deep neural network. Training is performed through a multi-layer residual connection network, so that the network output approximates the output of the reaction kinetics model. In this way, the deep neural network proxy model can replace complex reaction kinetics models for rapid calculations and predictions, improving computational efficiency.

[0069] Example 4:

[0070] This example details the specific implementation of the dynamic outcome prediction module. Based on a coupled kinetic model, the module uses a temporal convolutional network to predict product distribution probabilities under different experimental parameters and generate optimized reaction condition scenarios. Furthermore, the module constructs a bidirectional gated recurrent unit network to model long-term dependencies in the reaction process and extract multi-scale representations of temporal features. Monte Carlo sampling is used to generate a perturbation set of reaction parameters and calculate the probability density function of the product distribution.

[0071] When constructing a bidirectional gated recurrent unit network to model the long-term dependencies of the reaction process, the bidirectional gated recurrent unit (Bi-GRU) network consists of a forward GRU and a backward GRU. GRU is a gated recurrent unit that contains a reset gate. and update gate For the moment Input and the hidden state at the previous moment , reset gate , update gate ,in is the sigmoid function, , , , Is the weight matrix. Candidate hidden state , the final hidden state ,in Represents element-wise multiplication. The forward GRU processes the input sequentially starting from the beginning of the sequence, while the backward GRU processes the input backward starting from the end of the sequence. The outputs of the forward and backward GRUs are concatenated to form the output of a bidirectional gated recurrent unit network. This fully captures the long-term dependencies between different moments in the reaction process, providing a better understanding of the dynamics of the reaction.

[0072] When extracting multi-scale representations of temporal features, different convolution operations are used based on the bidirectional gated recurrent unit network to extract multi-scale temporal features. Convolution kernels of different sizes can be used, for example, , , The convolution kernel of the bidirectional gated recurrent unit network. , and perform convolution operations on them using convolution kernels of different sizes. Let the convolution kernel be , , , then the features after convolution , , ,in Represents a convolution operation. Convolution kernels of different sizes can capture features at different time scales. Small convolution kernels can capture detailed changes on short time scales, while large convolution kernels can capture trend changes on long time scales. This results in a multi-scale representation of temporal features and a more comprehensive analysis of the reaction process.

[0073] Monte Carlo sampling is a random sampling method that generates a set of perturbations of reaction parameters and calculates the probability density function of the product distribution. Assume that the reaction parameters are , respectively , ,..., First, determine the range of each reaction parameter , ,..., Then, a large number of random samplings are performed within these value ranges to generate sample points of reaction parameters, each of which is , , these sample points constitute the perturbation set of the reaction parameters. For each sample point, the corresponding product distribution is calculated using the coupled kinetic model. Let the product distribution be , through a large number of Perform statistical analysis and use methods such as kernel density estimation to calculate the probability density function of product distribution This can take into account the impact of uncertainty in reaction parameters on product distribution and more accurately predict the possibility of product distribution.

[0074] Example 5:

[0075] This example focuses on further expanding the capabilities of the dynamic outcome prediction module. In addition to the previously mentioned functions, the module also employs a generative adversarial network to construct a virtual experimental environment, generating simulated data for reaction processes under extreme conditions. Furthermore, a contrastive learning algorithm is used to align the feature distribution differences between the virtual data and the actual observed data.

[0076] When using a generative adversarial network to build a virtual experimental environment and generate simulation data of reaction processes under extreme conditions, the generative adversarial network (GAN) consists of a generator and the discriminator Composition. Generator The role of random noise is to Generate simulated data, i.e. , the simulated data generated here It is the reaction process data under extreme conditions, such as high temperature, high pressure, and other conditions that are difficult to easily achieve in actual experiments. The function of is to judge whether the input data is real actual observation data or simulated data generated by the generator. Its output is a probability value, which indicates the possibility that the input data is real data. During the training process, the generator and the discriminator conduct adversarial training. The generator attempts to generate more realistic simulated data, making it difficult for the discriminator to distinguish between true and false; the discriminator attempts to improve its own discrimination ability and accurately distinguish between real data and simulated data. Through continuous adversarial training, the generator can generate increasingly higher quality simulated data of reaction processes under extreme conditions. Assuming random noise Normal distribution ,in represents a vector with a mean of 0, Represents the unit covariance matrix. Generator It is a model composed of a multi-layer neural network that transforms random noise into As input, through a series of linear transformations and nonlinear activation functions, the simulated reaction process data is output Discriminator It is also a neural network that receives input data (which can be real data or simulated data generated by the generator), and after internal calculation, outputs a During training, the loss function of the generator is It aims to minimize the probability that the discriminator misclassifies the generated data as false data, which can be expressed as ; Loss function of the discriminator It aims to maximize the ability to correctly distinguish real data from generated data, which can be expressed as ,in Represents real observed data. By alternately optimizing the generator and the discriminator, the final generator can generate simulated data close to the reaction process under real extreme conditions.

[0077] When aligning the feature distribution differences between virtual data and actual observation data through contrastive learning algorithm, the core idea of ​​contrastive learning algorithm is to let the model learn the similarities and differences of data. First, feature extraction is performed on virtual data and actual observation data. Assume that the extracted virtual data features are , the actual observation data characteristics are Then, define a contrastive learning loss function. Common contrastive learning loss functions include InfoNCE loss function. The calculation formula of InfoNCE loss function is ,in is the number of data pairs, Representation characteristics and The similarity between them can usually be calculated using cosine similarity, i.e. , InfoNCE is a temperature hyperparameter used to adjust the distribution of similarity. This loss function aims to make the feature similarity between a pair of virtual data and actual observation data stand out among all other data pairs, thereby narrowing the difference in feature distribution between the virtual and actual data. During training, by continuously adjusting the model parameters to minimize the InfoNCE loss function, the model can better align the feature distributions of virtual and actual data. This allows for more effective integration of the actual situation when using virtual data for analysis and prediction, improving prediction accuracy and reliability.

[0078] Example 6:

[0079] This embodiment is described in detail around the experimental parameter feedback module. The experimental parameter feedback module, based on the reaction condition optimization scheme, adjusts the temperature gradient and stirring rate parameters of the reaction system in real time through an adaptive control algorithm, and also includes designing a fuzzy PID controller to adjust the power output of the thermostat, and dynamically adjusts the control parameters based on the gradient characteristics of the temperature distribution diagram. At the same time, a reinforcement learning framework is used to train the parameter adjustment strategy network, defining the state space as the multimodal feature vector of the current reaction system, and generating an optimal control instruction sequence by constraining the balance between energy consumption and reaction efficiency through a reward function.

[0080] When designing a fuzzy PID controller to adjust the power output of a thermostat, the fuzzy PID controller combines the advantages of fuzzy logic and PID control. The control law of the PID controller is: ,in is the output of the controller, i.e. the power output of the thermostat; is the proportionality coefficient, is the integration coefficient, is the differential coefficient; is the error, that is, the difference between the set temperature and the actual measured temperature. In the fuzzy PID controller, according to the gradient characteristics of the temperature distribution diagram, fuzzy logic is used to 、 and First, the gradient characteristics of the temperature distribution map are quantified and fuzzified, and converted into fuzzy language variables, such as "large", "medium", "small", etc. Then, according to the pre-set fuzzy rule base, the fuzzy reasoning is used to obtain 、 and The fuzzy rule base is established based on experience and experimental data. For example, when the temperature gradient is large and the temperature is higher than the set value, the appropriate increase To speed up the temperature adjustment and reduce To avoid integral saturation, increase To suppress excessive temperature fluctuations. Finally, the adjusted 、 and Substitute it into the PID control formula to calculate the power output of the constant temperature device and achieve precise control of the temperature of the reaction system.

[0081] When using the reinforcement learning framework to train the parameter adjustment policy network, reinforcement learning is a method that learns the optimal strategy by interacting with the environment and based on the reward signal. In this system, the state space is defined as the multimodal feature vector of the current reaction system, including information such as reactant concentration, temperature, and pressure, denoted as The action space is the temperature gradient of the reaction system and the adjustment value of the stirring rate, which is recorded as The reward function is used to constrain the balance between energy consumption and reaction efficiency. Let the reward function be For example, when the reaction efficiency is improved and the energy consumption is low, a higher reward is given; when the reaction efficiency is reduced or the energy consumption is too high, a lower reward is given. Parameter adjustment strategy network According to the current status Output an action The probability distribution of During the training process, the agent Observe the current status , select an action according to the policy network And execute, then observe the new state and get rewards By continuously interacting with the environment, we accumulate experience data , use this data to update the policy network , so that the policy network can learn the optimal strategy to maximize the cumulative reward. Common reinforcement learning algorithms such as the proximal policy optimization algorithm (PPO) update the policy network by optimizing an objective function. The objective function is ,in are the parameters of the policy network, is the parameter of the last update, is the sample size, is the advantage function, indicating taking action In state The degree of advantage under The function will The value is limited to Within the range, is a hyperparameter. Through continuous iterative training, the policy network can generate the optimal control instruction sequence to achieve optimal adjustment of the temperature gradient and stirring rate of the reaction system, improve reaction efficiency and reduce energy consumption.

[0082] 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.

[0083] 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-driven chemical experiment simulation and result prediction system, characterized by: include: Experimental data acquisition module: used to collect multimodal data of the experimental process through a chemical sensor array and a multispectral imaging device. The multimodal data includes reactant concentration time series data, solution temperature distribution images, gas diffusion dynamic video streams, and reaction vessel pressure waveforms; Spectral feature analysis module: uses wavelet transform combined with convolutional neural network to extract spectral features of the multimodal data, generating multi-dimensional analysis results including material absorption peak position characteristics, reaction rate correlation characteristics and phase change process spatiotemporal characteristics; Reaction kinetics modeling module: Based on the multi-dimensional analytical results, a coupled kinetic model of the reactant concentration field and the temperature field is constructed using partial differential equations, and the gradient change of the reaction path is solved by the finite element method; Dynamic result prediction module: Based on the coupled kinetic model, a time-series convolutional network is used to predict the product distribution probability under different experimental parameters and generate a reaction condition optimization plan; Experimental parameter feedback module: According to the reaction condition optimization plan, the temperature gradient and stirring rate parameters of the reaction system are adjusted in real time through an adaptive control algorithm.

2. The chemical experiment simulation and result prediction system according to claim 1, characterized in that: The experimental data acquisition module includes: An electrochemical sensor array was used to measure the time series data of ion mobility at the reaction interface, and an infrared thermal imager was used to capture the surface temperature distribution of the reaction vessel. A high-speed camera was used to record the turbulent characteristic video stream of the gas evolution process, and the oscillation waveform data of the pressure sensor in the reactor was collected synchronously.

3. The chemical experiment simulation and result prediction system according to claim 2, characterized in that: The spectral feature analysis module also includes: A frequency domain decomposition algorithm is used to separate the spectral responses of different bands of the multimodal data, and a weighted fusion feature channel is performed in combination with an attention mechanism; The vortex structure characteristics of the gas diffusion dynamic video flow are extracted by a three-dimensional convolution kernel, and a correlation matrix between turbulence intensity and reaction rate is constructed.

4. The chemical experiment simulation and result prediction system according to claim 1, characterized in that: The reaction kinetics modeling module also includes: Establish a non-steady-state mass transfer equation to describe the concentration gradient of reactants, and input the temperature field data as the boundary condition; A multi-grid iterative algorithm is used to accelerate the partial differential equation solution process, and a deep neural network proxy model is constructed based on residual connections.

5. The chemical experiment simulation and result prediction system according to claim 1, characterized in that: The dynamic result prediction module also includes: A bidirectional gated recurrent unit network is constructed to model the long-term dependencies of the reaction process, extract multi-scale representations of temporal features, generate a perturbation set of reaction parameters through Monte Carlo sampling, and calculate the probability density function of the product distribution.

6. The chemical experiment simulation and result prediction system according to claim 5, characterized in that: The dynamic result prediction module also includes: Generative adversarial networks are used to construct a virtual experimental environment to generate simulation data of reaction processes under extreme conditions. The feature distribution differences between virtual data and actual observed data are aligned through contrastive learning algorithm.

7. The chemical experiment simulation and result prediction system according to claim 1, characterized in that: The experimental parameter feedback module also includes: designing a fuzzy PID controller to adjust the power output of the constant temperature device, and dynamically adjusting the control parameters based on the gradient characteristics of the temperature distribution diagram.

8. The chemical experiment simulation and result prediction system according to claim 7, characterized in that: The experimental parameter feedback module also includes: A reinforcement learning framework is used to train the parameter adjustment strategy network, defining the state space as the multimodal feature vector of the current reaction system; The reward function constrains the balance between energy consumption and reaction efficiency to generate the optimal control instruction sequence.

9. The chemical experiment simulation and result prediction system according to claim 1, characterized in that: The system further comprises: A reaction safety assessment model is constructed, and based on the gradient field data of the coupled dynamics model, a support vector machine classifier is used to identify the reaction runaway risk threshold; when risk features are detected, an emergency parameter adjustment protocol is triggered and a safety protection recommendation plan is generated.

10. The chemical experiment simulation and result prediction system according to claim 9, characterized in that: The reaction safety assessment model also includes: associating historical accident case data through a knowledge graph to generate a safety protection strategy reasoning path for the current reaction system.

Images