A method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction

Abstract

The invention provides a parameter determination method for improving the efficiency of methyl oleate epoxidation reaction, which belongs to the technical field of methyl oleate epoxidation reaction, and comprises: collecting multiple groups of small-scale test data of methyl oleate epoxidation reaction; analyzing the small-scale test data by using a Bayesian optimization algorithm to obtain the correlation coefficient between each reaction condition parameter and the reaction efficiency data, and clustering according to the correlation coefficient to obtain key parameters, secondary parameters and micro-influence parameters; establishing a training data set, and training to obtain a two-way model of methyl oleate epoxidation reaction; discretizing a preset reaction efficiency requirement range, and inputting it into the model to obtain a reaction condition parameter set; using a genetic algorithm to use a to-be-selected reaction condition parameter set as an initial population, and iteratively optimizing to obtain target reaction condition parameters that meet the requirements; screening out a seed reaction condition parameter set; designing an orthogonal experiment to obtain an experimental data set, and selecting the reaction efficiency that is most similar to the preset target as the optimal reaction condition parameter.

Description

Technical Field

[0001] The invention belongs to the technical field of methyl oleate epoxidation reaction, and in particular relates to a parameter determination method for improving the efficiency of methyl oleate epoxidation reaction. Background Art

[0002] Methyl oleate is a common fatty acid derivative, widely used in food additives, cosmetics, lubricants and other fields. Among them, the epoxidation reaction of methyl oleate is an important derivative reaction, which can obtain epoxy methyl oleate products with higher physical and chemical properties and biological activity. The efficiency of the epoxidation reaction directly affects the yield and quality of the product, so how to improve the efficiency of the epoxidation reaction of methyl oleate has always been the research focus in related fields.

[0003] At present, the common methods for methyl oleate epoxidation reaction mainly include peroxide method, halogen oxidation method and transition metal catalysis method. Among them, the transition metal catalysis method has attracted widespread attention due to its advantages such as mild reaction conditions and good selectivity. Commonly used catalysts include transition metal compounds such as vanadium, molybdenum, and manganese. However, these reaction systems usually involve a variety of reaction condition parameters, such as reaction temperature, catalyst dosage, oxidant dosage, reaction time, etc. There are mutual coupling and complex influence relationships between the parameters, and it is difficult to optimize a single parameter to achieve satisfactory reaction efficiency. In addition, most of the existing optimization methods use single-factor experimental methods or orthogonal experimental methods, which are limited to local searches in the parameter space. It is difficult to fully explore the influence of parameters on reaction efficiency and it is difficult to obtain the global optimal reaction conditions. Summary of the invention

[0004] In view of this, the present invention provides a parameter determination method for improving the efficiency of methyl oleate epoxidation reaction, which can solve the technical problem that most existing optimization methods adopt single-factor experimental methods or orthogonal experimental methods, are limited to local search in parameter space, and are difficult to comprehensively explore the influence of parameters on reaction efficiency and difficult to obtain the global optimal reaction conditions.

[0005] The present invention is achieved in that:

[0006] The present invention provides a method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction, which comprises the following steps:

[0007] S10, collecting multiple groups of small-scale test data of methyl oleate epoxidation reaction, including reaction condition parameters and reaction efficiency data;

[0008] S20, analyzing the pilot test data using a Bayesian optimization algorithm to obtain a correlation coefficient between each reaction condition parameter and the reaction efficiency data, and clustering the correlation coefficients to obtain three types of reaction condition parameters, specifically key parameters, secondary parameters, and micro-influence parameters;

[0009] S30, establishing a training data set according to the multiple groups of small test data, training a bidirectional neural network model with a multi-branch tree structure, and obtaining a bidirectional model for the epoxidation reaction of methyl oleate; wherein the input of the training is the reaction condition parameter, and the output of the training is the reaction efficiency data;

[0010] S40, discretizing the preset reaction efficiency requirement range to obtain multiple reaction efficiency requirements, inputting each reaction efficiency requirement into the bidirectional model of the methyl oleate epoxidation reaction to obtain corresponding reaction condition parameters, and merging the reaction condition parameters corresponding to each reaction efficiency requirement into a set of reaction condition parameters to be selected;

[0011] S50, using a genetic algorithm, taking the set of reaction condition parameters to be selected as the initial population, taking the reaction efficiency data as the fitness function, setting the crossover function as arithmetic crossover, and the mutation function as weighted non-uniform mutation, performing iterative optimization, and finally obtaining the target reaction condition parameters that meet the iteration stop condition;

[0012] S60, among the intermediate reaction condition parameters outputted in each iteration of the genetic algorithm, selecting the intermediate reaction condition parameters whose weighted similarity with the target reaction condition parameters is greater than the weighted similarity threshold as a screening set, and merging the screening set and the target reaction condition parameters into a seed reaction condition parameter set;

[0013] S70, designing an orthogonal experiment according to the seed reaction condition parameter set with the correlation coefficient of each reaction condition parameter as an importance factor, and obtaining an experimental data set of the orthogonal experiment, wherein each experimental data in the experimental data set includes reaction condition parameters and reaction efficiency data;

[0014] S80. In the experimental data set, the reaction condition parameters of the experimental data whose reaction efficiency data has the highest similarity with the preset target reaction efficiency data are selected as the optimal reaction condition parameters.

[0015] Wherein, the reaction efficiency data include product yield, selectivity and conversion rate.

[0016] The reaction condition parameters include reaction temperature, reaction time, catalyst dosage, oxidant dosage, stirring speed, reaction pressure, raw material concentration, pH value, solvent type, solvent dosage, reactor type, reaction material ratio, additive type, and additive dosage.

[0017] The weight of the weighted similarity is the correlation coefficient corresponding to each reaction condition parameter.

[0018] The variation function is a weighted non-uniform variation, specifically, a weighted non-uniform variation is performed using the correlation coefficient of the reaction condition parameters as a weight.

[0019] Among them, the bidirectional neural network model includes a key parameter subnetwork, a secondary parameter subnetwork, a micro-influence parameter subnetwork, an overall evaluation subnetwork and a fusion subnetwork, and the tree structure is specifically: the outputs of the key parameter subnetwork, the secondary parameter subnetwork, the micro-influence parameter subnetwork and the overall evaluation subnetwork are the inputs of the fusion subnetwork.

[0020] Furthermore, the key parameter sub-network is used to process key parameters, the training input is the key parameters, the training output is the reaction efficiency data, and its structure is a multi-layer perceptron network;

[0021] The secondary parameter subnetwork is used to process secondary parameters, the training input is the secondary parameters, the training output is the reaction efficiency data, and its structure is a convolutional neural network;

[0022] The micro-influence parameter sub-network is used to process micro-influence parameters, the training input is the micro-influence parameters, the training output is the reaction efficiency data, and its structure is a recurrent neural network.

[0023] Furthermore, the key parameter is a parameter with a correlation coefficient greater than 0.7; the secondary parameter is a parameter with a correlation coefficient greater than or equal to 0.4 and less than or equal to 0.7; and the micro-influence parameter is a parameter with a correlation coefficient less than 0.4.

[0024] Furthermore, the overall evaluation sub-network is used to adopt a self-attention mechanism and global information integration, the training input is all reaction condition parameters, the training output is reaction efficiency data, and its structure is a Transformer network.

[0025] Furthermore, the fusion sub-network is used to adopt a counterfactual reasoning mechanism and multimodal information fusion. The training input is all reaction condition parameters and the output of each sub-network, and the training output is reaction efficiency data. Its structure is a graph neural network.

[0026] Specifically, step S10 includes the following steps: determining the main reaction condition parameters that affect the efficiency of the methyl oleate epoxidation reaction, including reaction temperature, reaction time, catalyst dosage, oxidant dosage, stirring speed, reaction pressure, raw material concentration, pH value, solvent type, solvent dosage, reactor type, reaction material ratio, additive type and additive dosage, etc.; designing and conducting multiple groups of small-scale experiments to collect methyl oleate epoxidation reaction efficiency data under different reaction condition parameter combinations, including product yield, selectivity and conversion rate; organizing the collected small-scale test data into a standard data table format to provide a data basis for subsequent analysis and model training.

[0027] Wherein, the step S20 includes the following steps: using the Bayesian optimization algorithm to analyze the pilot test data to obtain the correlation coefficient between each reaction condition parameter and the reaction efficiency data; according to the correlation coefficient of each parameter, the reaction condition parameters are divided into three categories: key parameters, secondary parameters and micro-influence parameters; the division results of these three types of parameters are recorded to provide a basis for subsequent model training and parameter optimization.

[0028] Wherein, the step S30 includes the following steps: designing a multi-branch tree-structured bidirectional neural network model, including a key parameter subnetwork, a secondary parameter subnetwork, a micro-influence parameter subnetwork, an overall evaluation subnetwork and a fusion subnetwork; wherein the key parameter subnetwork adopts a multi-layer perceptron network structure, the secondary parameter subnetwork adopts a convolutional neural network structure, the micro-influence parameter subnetwork adopts a recurrent neural network structure, the overall evaluation subnetwork adopts a Transformer network structure, and the fusion subnetwork adopts a graph neural network structure; using the pilot test data as a training set, training the bidirectional neural network model so that it can accurately predict the reaction efficiency.

[0029] Wherein, the step S40 includes the following steps: according to actual production requirements, determining the range of target reaction efficiency, for example, product yield not less than 85%, selectivity not less than 90%, and conversion rate not less than 95%; discretizing these reaction efficiency requirements to obtain 27 different reaction efficiency requirement combinations; inputting each reaction efficiency requirement into a bidirectional neural network model to obtain corresponding reaction condition parameters, and combining all output reaction condition parameters into a set of candidate reaction condition parameters.

[0030] Wherein, the step S50 includes the following steps: using the candidate reaction condition parameter set generated in step S40 as the initial population of the genetic algorithm; designing the fitness function of the genetic algorithm as the weighted sum of the reaction efficiency data; setting the crossover function to arithmetic crossover and the mutation function to weighted non-uniform mutation; performing multiple rounds of genetic algorithm iterative optimization until the preset stop condition is met, and obtaining the optimal reaction condition parameters that meet the target reaction efficiency requirements. Wherein, for the fitness function in the genetic algorithm, the weighted sum of the reaction efficiency data can be used:

[0031] ;

[0032] In the formula, is the fitness function value; is the product yield; is selective; is the conversion rate; , , is the weight coefficient, satisfying ;The weight coefficient can be adjusted according to specific needs, for example, it can be determined through expert opinion or historical data analysis.

[0033] Weighted non-uniform mutation formula of genetic algorithm:

[0034]

[0035] In the formula, is the first parameter values; is the first parameter values; is the current iteration number; It is The upper limit of the parameters; It is The correlation coefficient of the parameters;

[0036] The function is defined as:

[0037] ;

[0038] In the formula, is the maximum number of iterations; is a A random number between ; is a shape parameter, usually taking a value of 2 or 3.

[0039] Wherein, the step S60 includes the following steps: defining a weighted similarity index for evaluating the similarity between the intermediate reaction condition parameters and the target reaction condition parameters; in each iteration of the genetic algorithm, selecting the intermediate reaction condition parameters whose weighted similarity is greater than a preset similarity threshold (e.g., 0.9) as a screening set; merging the screening set and the target reaction condition parameters obtained in step S50 to form a seed reaction condition parameter set.

[0040] Among them, the weighted similarity calculation formula is:

[0041] ;

[0042] In the formula, is the weighted similarity; It is The correlation coefficient of the parameters; is the intermediate reaction condition parameter Values; is the target reaction condition parameter Values; and They are Upper and lower limits for each parameter; is the total number of parameters.

[0043] Wherein, the step S70 includes the following steps: based on the seed reaction condition parameter set obtained in step S60, an orthogonal test scheme is designed according to the correlation coefficient of each parameter as an importance factor; experiments are conducted according to the orthogonal test scheme to obtain an experimental data set; in the experimental data set, the experimental data whose reaction efficiency data has the highest similarity with the preset target reaction efficiency data is selected as the optimal reaction condition parameter.

[0044] In step S80, the reaction efficiency data similarity calculation formula is:

[0045] ;

[0046] In the formula, is the reaction efficiency data similarity; , , are the product yield, selectivity and conversion rate of the experimental data respectively; , , They are the product yield, selectivity and conversion rate of the target reaction efficiency data. This formula calculates the normalized Euclidean distance, and the result range is The larger the value, the higher the similarity.

[0047] Compared with the prior art, the method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction provided by the present invention has the following beneficial effects:

[0048] 1. The Bayesian optimization algorithm is used to analyze the test data, which can comprehensively determine the influence of each reaction condition parameter on the reaction efficiency, and divide the parameters into key parameters, secondary parameters and micro-influence parameters, providing a basis for subsequent model design and parameter optimization. Compared with the traditional single factor experimental method or orthogonal experimental method, this method can better explore the mutual coupling relationship between parameters.

[0049] 2. A bidirectional neural network model with a multi-branch tree structure was established, which can effectively capture the influence mechanism of different types of parameters on reaction efficiency. The multi-branch structure of the model can make full use of the difference in parameter importance and improve the prediction accuracy; the bidirectional structure can realize the bidirectional mapping between parameters and reaction efficiency, providing a basis for subsequent parameter optimization.

[0050] 3. The genetic algorithm is used to perform global optimization on the parameter set to be selected, and the optimal parameter combination that meets the target reaction efficiency requirements can be quickly found. The genetic algorithm has good global search capabilities and can avoid falling into local optimality. In particular, the present invention uses a genetic algorithm with weighted non-uniform mutation, which can better utilize parameter importance information and improve convergence speed and solution quality.

[0051] 4. Through weighted similarity screening and orthogonal test verification, the search range of parameter optimization can be narrowed and the reliability of the final solution can be improved. The weighted similarity index fully considers the importance of parameters and can effectively retain the intermediate solutions corresponding to key parameters; orthogonal tests can explore the parameter space within a limited number of experiments and find the best parameter combination.

[0052] In summary, the parameter determination method for improving the epoxidation reaction efficiency of methyl oleate proposed in the present invention systematically determines the optimal reaction condition parameters that meet the target reaction efficiency requirements by combining data-driven analysis, model prediction and optimization algorithm. Compared with the existing technology, it has higher reaction efficiency and stronger reliability, and solves the technical problem that most existing optimization methods use single-factor experimental methods or orthogonal experimental methods, which are limited to local searches in parameter space, making it difficult to comprehensively explore the influence of parameters on reaction efficiency and difficult to obtain the global optimal reaction conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0055] like Figure 1 FIG. 1 is a flow chart of a method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction provided by the present invention. The method comprises the following steps:

[0056] S10, collecting multiple groups of small-scale test data of methyl oleate epoxidation reaction, including reaction condition parameters and reaction efficiency data;

[0057] S20, using the Bayesian optimization algorithm to analyze the pilot test data, obtain the correlation coefficient between each reaction condition parameter and the reaction efficiency data, and cluster them according to the correlation coefficient to obtain three types of reaction condition parameters, specifically key parameters, secondary parameters and micro-influence parameters;

[0058] S30, establishing a training data set based on multiple groups of small test data, training a bidirectional neural network model with a multi-branch tree structure, and obtaining a bidirectional model for the epoxidation reaction of methyl oleate; wherein the input of the training is the reaction condition parameter, and the output of the training is the reaction efficiency data;

[0059] S40, discretizing the preset reaction efficiency requirement range to obtain multiple reaction efficiency requirements, inputting each reaction efficiency requirement into the bidirectional model of methyl oleate epoxidation reaction to obtain corresponding reaction condition parameters, and merging the reaction condition parameters corresponding to each reaction efficiency requirement into a set of reaction condition parameters to be selected;

[0060] S50, using a genetic algorithm, taking the reaction condition parameter set to be selected as the initial population, taking the reaction efficiency data as the fitness function, setting the crossover function as arithmetic crossover, and the mutation function as weighted non-uniform mutation, performing iterative optimization, and finally obtaining the target reaction condition parameters that meet the iteration stop condition;

[0061] S60, among the intermediate reaction condition parameters outputted in each iteration of the genetic algorithm, selecting the intermediate reaction condition parameters whose weighted similarity with the target reaction condition parameters is greater than the weighted similarity threshold as a screening set, and merging the screening set and the target reaction condition parameters into a seed reaction condition parameter set;

[0062] S70, designing an orthogonal experiment according to the seed reaction condition parameter set with the correlation coefficient of each reaction condition parameter as an importance factor, and obtaining an experimental data set of the orthogonal experiment, wherein each experimental data in the experimental data set includes reaction condition parameters and reaction efficiency data;

[0063] S80. In the experimental data set, the reaction condition parameters of the experimental data whose reaction efficiency data has the highest similarity with the preset target reaction efficiency data are selected as the optimal reaction condition parameters.

[0064] The specific implementation methods of the above steps are described in detail below:

[0065] Step S10: Collecting multiple groups of small test data of methyl oleate epoxidation reaction

[0066] The specific implementation process of this step is as follows:

[0067] 1) According to the existing literature and research experience, the main reaction condition parameters affecting the efficiency of methyl oleate epoxidation reaction were determined, including reaction temperature, reaction time, catalyst dosage, oxidant dosage, stirring speed, reaction pressure, raw material concentration, pH value, solvent type, solvent dosage, reactor type, reaction material ratio, additive type and additive dosage, etc.

[0068] 2) Design and conduct multiple groups of small-scale experiments to collect the efficiency data of oleic acid methyl ester epoxidation under different reaction condition parameter combinations, including product yield, selectivity and conversion rate. Each group of small-scale experimental data should contain complete reaction condition parameters and corresponding reaction efficiency index data.

[0069] 3) Organize the collected small-scale test data into a standard data table format for subsequent analysis and model training.

[0070] The role of this step is to provide the necessary experimental data basis for subsequent parameter optimization and reaction condition determination. Through small-scale experiments, we can comprehensively collect the impact of reaction condition parameters on reaction efficiency, laying a data foundation for the next step of analysis and optimization.

[0071] Step S20: Using Bayesian optimization algorithm to analyze the pilot test data and determine the correlation and classification of reaction condition parameters

[0072] The specific implementation process of this step is as follows:

[0073] 1) Use the Bayesian optimization algorithm to analyze the collected small test data. The Bayesian optimization algorithm is an optimization method based on Gaussian processes, which can effectively use historical experimental data and quickly find the optimal solution with a small number of experiments.

[0074] 2) Through the analysis of the Bayesian optimization algorithm, the correlation coefficient between each reaction condition parameter and the reaction efficiency data can be obtained. The higher the correlation coefficient, the greater the impact of the parameter on the reaction efficiency.

[0075] 3) According to the correlation coefficient of each parameter, the reaction condition parameters are divided into three categories:

[0076] Key parameters: Parameters with higher correlation coefficients have a greater impact on reaction efficiency.

[0077] Secondary parameters: Parameters with medium correlation coefficients have a certain influence on the reaction efficiency.

[0078] Micro-influence parameters: Parameters with lower correlation coefficients have less impact on reaction efficiency.

[0079] 4) Record the division results of these three types of parameters to provide a basis for subsequent model training and parameter optimization.

[0080] The role of this step is to conduct an in-depth analysis of the experimental data through the Bayesian optimization algorithm, determine the degree of influence of each reaction condition parameter on the reaction efficiency, and provide a basis for parameter classification for subsequent steps. Dividing parameters into key parameters, secondary parameters, and micro-influence parameters helps to better design model structure and optimization algorithm.

[0081] Step S30: Establishing a bidirectional neural network model for methyl oleate epoxidation reaction

[0082] The specific implementation process of this step is as follows:

[0083] 1) According to the three types of parameter features determined in step S20, a multi-branch tree-structured bidirectional neural network model is designed. The model includes the following sub-networks:

[0084] Key parameter sub-network: uses a multi-layer perceptron network structure to process the impact of key parameters on reaction efficiency;

[0085] Secondary parameter sub-network: uses a convolutional neural network structure to process the influence of secondary parameters on reaction efficiency;

[0086] Micro-influence parameter sub-network: adopts a recurrent neural network structure to process the influence of micro-influence parameters on reaction efficiency;

[0087] Overall evaluation sub-network: adopts Transformer network structure, uses self-attention mechanism and global information integration to comprehensively evaluate the impact of all parameters on reaction efficiency;

[0088] Fusion sub-network: Using a graph neural network structure, utilizing counterfactual reasoning mechanism and multimodal information fusion, the outputs of each sub-network are comprehensively integrated to obtain the final reaction efficiency prediction.

[0089] 2) Using the small test data collected in step S10 as a training set, the established bidirectional neural network model is trained, wherein the input of the model is the reaction condition parameters, and the output is the corresponding reaction efficiency data.

[0090] 3) During the training process, the back propagation algorithm is used to optimize the model parameters with the goal of minimizing the error between the predicted reaction efficiency and the actual reaction efficiency.

[0091] The purpose of this step is to establish a bidirectional neural network model covering different types of parameters, which can accurately predict the efficiency of methyl oleate epoxidation reaction. The multi-branch tree structure of the model can effectively capture the influence mechanism of different types of parameters on reaction efficiency and improve the prediction accuracy of the model.

[0092] Step S40: Generate a set of reaction condition parameters to be selected according to the reaction efficiency requirements

[0093] The specific implementation process of this step is as follows:

[0094] 1) Determine the range of target reaction efficiency based on actual production needs, such as product yield not less than 85%, selectivity not less than 90%, and conversion rate not less than 95%.

[0095] 2) Appropriately discretize these reaction efficiency requirements, for example, divide the yield into three levels of 85%, 87%, and 90%, the selectivity into three levels of 90%, 92%, and 95%, and the conversion rate into three levels of 95%, 97%, and 99%. In this way, 27 (3x3x3) different reaction efficiency requirement combinations are obtained.

[0096] 3) Input each reaction efficiency requirement combination into the bidirectional neural network model established in step S30 to obtain the corresponding reaction condition parameters.

[0097] 4) Combine all output reaction condition parameters into a candidate reaction condition parameter set to lay the foundation for subsequent optimization.

[0098] The purpose of this step is to generate a set of candidate reaction condition parameters that meet the requirements of various reaction efficiency indicators based on actual production needs. Using the established bidirectional neural network model, parameter configuration solutions that meet different efficiency requirements can be quickly obtained, providing sufficient parameter candidate space for subsequent optimization.

[0099] Step S50: Optimizing the reaction condition parameter set to be selected using a genetic algorithm

[0100] The specific implementation process of this step is as follows:

[0101] 1) The reaction condition parameter set to be selected generated in step S40 is used as the initial population of the genetic algorithm.

[0102] 2) Design the fitness function of the genetic algorithm as the weighted sum of the reaction efficiency data, that is:

[0103] ;

[0104] in, is the fitness function value; is the product yield; is selective; is the conversion rate; , , It is the weight coefficient set according to actual needs, satisfying .

[0105] 3) Set the crossover function of the genetic algorithm to arithmetic crossover and the mutation function to weighted non-uniform mutation. The weighted non-uniform mutation formula is:

[0106] ;

[0107] in, is the first parameter values; is the first parameter values; is the current iteration number; It is The upper limit of the parameters; It is The correlation coefficient of the parameters; is a function that controls the amplitude of variation and is defined as:

[0108] ;

[0109] in, is the maximum number of iterations; is a A random number between ; is the shape parameter, usually 2 or 3.

[0110] 4) Perform multiple rounds of iterative optimization using genetic algorithms until the preset stopping conditions are met, and obtain the optimal reaction condition parameters that meet the target reaction efficiency requirements.

[0111] The purpose of this step is to use the genetic algorithm to iteratively optimize the candidate parameter set generated in step S40 to obtain the optimal reaction condition parameter combination that meets the target reaction efficiency requirements. The genetic algorithm is an effective global optimization algorithm that can quickly find a parameter configuration close to the optimal solution through fitness function evaluation and genetic operations. In particular, the design of weighted non-uniform mutation can better utilize the correlation information of parameters and improve the optimization efficiency.

[0112] Step S60: Screening intermediate reaction condition parameters based on weighted similarity

[0113] The specific implementation process of this step is as follows:

[0114] 1) Define weighted similarity index , used to evaluate the similarity between the intermediate reaction condition parameters and the target reaction condition parameters:

[0115] ;

[0116] in, is the weighted similarity; It is The correlation coefficient of the parameters; is the intermediate reaction condition parameter Values; is the target reaction condition parameter Values; and They are The upper and lower limits of the parameters; is the total number of parameters.

[0117] 2) In each iteration of the genetic algorithm, select the weighted similarity The intermediate reaction condition parameters that are greater than a preset similarity threshold (e.g., 0.9) are used as a screening set.

[0118] 3) The screening set and the target reaction condition parameters obtained in step S50 are combined to form a seed reaction condition parameter set.

[0119] The role of this step is to select parameter combinations that are more similar to the target reaction condition parameters from the intermediate results of the genetic algorithm iteration through the weighted similarity index, and use them as the seed parameter set for the subsequent orthogonal test. This can minimize the scope of parameter optimization while ensuring the reaction efficiency, thereby improving the efficiency of the orthogonal test.

[0120] Step S70: Optimizing reaction condition parameters based on orthogonal experimental design

[0121] The specific implementation process of this step is as follows:

[0122] 1) Based on the seed reaction condition parameter set obtained in step S60, an orthogonal test scheme is designed according to the correlation coefficient of each parameter as an importance factor.

[0123] 2) Conduct experiments according to the orthogonal experimental scheme to obtain experimental data sets. Each data item in the experimental data set contains reaction condition parameters and corresponding reaction efficiency data.

[0124] 3) In the experimental data set, select the experimental data whose reaction efficiency data is most similar to the preset target reaction efficiency data, and the corresponding reaction condition parameters are the optimal reaction condition parameters.

[0125] Reaction efficiency data similarity The calculation formula is:

[0126] ;

[0127] in, is the reaction efficiency data similarity; , , are the product yield, selectivity and conversion rate of the experimental data respectively; , , They are the product yield, selectivity and conversion of the target reaction efficiency data respectively; , , is the weight coefficient of the corresponding indicator.

[0128] The purpose of this step is to further optimize and verify the seed parameter set obtained in step S60 by means of orthogonal experiments. Orthogonal experiments can fully explore the parameter space within a limited number of experiments and find the optimal parameter combination that is most similar to the target reaction efficiency. As an importance factor, the correlation coefficient can better focus on the key parameters that have a greater impact on the reaction efficiency.

[0129] Step S80: Determine the final optimal reaction condition parameters

[0130] The specific implementation process of this step is as follows:

[0131] 1) In the orthogonal test data set of step S70, find the experimental data whose reaction efficiency data has the highest similarity with the preset target reaction efficiency data.

[0132] 2) The reaction condition parameters corresponding to the experimental data are determined as the final optimal reaction condition parameters.

[0133] The purpose of this step is to determine the optimal reaction condition parameter combination that meets the target reaction efficiency requirements based on the optimization results of the previous steps. Through a series of analysis, modeling and optimization, the advantages of experimental data and model prediction are fully utilized to finally determine the reaction condition parameters that can best meet production needs.

[0134] In summary, this parameter determination method for improving the efficiency of methyl oleate epoxidation reaction, through steps such as small-scale experimental data collection, Bayesian algorithm parameter classification, bidirectional neural network model establishment, genetic algorithm optimization, weighted similarity screening and orthogonal test verification, systematically analyzes and optimizes various parameters that affect the reaction efficiency, and finally determines the optimal reaction condition parameters that meet production requirements.

[0135] Specifically, the principle of the present invention is:

[0136] 1. Collecting small-scale experimental data: First, through small-scale experiments, systematically collect the efficiency data of oleic acid methyl ester epoxidation reaction under different reaction condition parameter combinations, including product yield, selectivity and conversion rate. These experimental data provide the necessary basis for subsequent analysis and optimization.

[0137] 2. Parameter importance analysis: The Bayesian optimization algorithm is used to analyze the pilot test data to obtain the correlation coefficient between each reaction condition parameter and the reaction efficiency. According to the correlation coefficient of the parameters, they are divided into three categories: key parameters, secondary parameters, and minor parameters. This parameter importance analysis provides a basis for subsequent model design and optimization algorithms, and helps to better focus on the key parameters that have a greater impact on reaction efficiency.

[0138] 3. Bidirectional neural network modeling: Based on the three types of parameter features determined in step 2, the present invention designs a bidirectional neural network model with a multi-branch tree structure. The model includes a key parameter subnetwork, a secondary parameter subnetwork, a micro-influence parameter subnetwork, an overall evaluation subnetwork, and a fusion subnetwork. Different subnetworks use different neural network structures, such as multilayer perceptron networks, convolutional neural networks, recurrent neural networks, etc., to better capture the influence mechanism of different types of parameters on reaction efficiency. The bidirectional structure can realize the bidirectional mapping between parameters and reaction efficiency, providing a basis for subsequent parameter optimization.

[0139] 4. Generation of candidate parameters: According to actual production requirements, determine the target reaction efficiency indicators, such as product yield, selectivity and conversion rate. Discretize these target efficiency indicators to obtain multiple reaction efficiency requirements. Input each requirement into the bidirectional neural network model established in step 3 to quickly obtain the reaction condition parameter combination that meets the requirement, thereby generating a candidate parameter set.

[0140] 5. Parameter optimization: Genetic algorithm is used to globally optimize the candidate parameter set generated in step 4. Genetic algorithm is a random optimization algorithm based on natural selection and genetic mechanism, and has good global search ability. The present invention adopts weighted non-uniform mutation in genetic algorithm, which can better utilize parameter importance information and improve the convergence of the algorithm and the quality of the solution.

[0141] 6. Parameter screening and verification: During the iteration process of the genetic algorithm, the weighted similarity index is used to screen the intermediate results, and only the solutions that are similar to the target parameters are retained. This step can effectively narrow the search space for parameter optimization. Finally, the orthogonal test method is used to further optimize and verify the screened parameter set, and the optimal parameter combination that meets the target reaction efficiency requirements can be found within a limited number of experiments.

[0142] In summary, the parameter determination method proposed in the present invention fully combines technical means such as data analysis, model building and optimization algorithm, and has obvious advantages in systematically exploring parameter space, accurately predicting reaction efficiency, and ultimately determining the optimal parameter combination that meets target requirements.

[0143] In order to better understand and implement the present invention, the following provides an embodiment 1 of a specific application scenario of the present invention: The steps of this embodiment 1 are specifically described as follows:

[0144] Step S10: Collecting multiple groups of small test data of methyl oleate epoxidation reaction

[0145] The specific implementation process of this step is as follows:

[0146] 1) Determine the main reaction parameters that affect the efficiency of methyl oleate epoxidation, including reaction temperature , reaction time , Catalyst dosage , Oxidant dosage , stirring speed , reaction pressure , Raw material concentration , pH value , Solvent type , Solvent dosage , Reactor type , reaction material ratio , Additive Type And the amount of additives wait.

[0147] 2) Design and conduct multiple small-scale experiments to collect data on the epoxidation efficiency of methyl oleate under different reaction conditions and parameter combinations, including product yield. , Selectivity and conversion rate Each set of pilot test data should contain complete reaction condition parameters and corresponding reaction efficiency index data, which can be expressed as a vector and the corresponding reaction efficiency index vector .

[0148] 3) Organize the collected small-scale test data into a standard data table format for subsequent analysis and model training.

[0149] The role of this step is to provide the necessary experimental data basis for subsequent parameter optimization and reaction condition determination. Through small-scale experiments, we can comprehensively collect the impact of reaction condition parameters on reaction efficiency, laying a data foundation for the next step of analysis and optimization.

[0150] Step S20: Using Bayesian optimization algorithm to analyze the pilot test data and determine the correlation and classification of reaction condition parameters

[0151] The specific implementation process of this step is as follows:

[0152] 1) Use the Bayesian optimization algorithm to analyze the collected small test data. The Bayesian optimization algorithm is an optimization method based on Gaussian process, which can effectively use historical experimental data and quickly find the optimal solution with a small number of experiments. Its basic idea is:

[0153]

[0154] in, is the reaction efficiency index vector, is the reaction condition parameter vector, is the unknown objective function, is a noise term that obeys a normal distribution. Through Bayesian inference, we can get the objective function The posterior distribution of is used to determine the optimal parameter combination.

[0155] 2) Through the analysis of Bayesian optimization algorithm, the parameters of each reaction condition can be obtained Reaction efficiency data The correlation coefficient between The higher the correlation coefficient, the greater the effect of the parameter on the reaction efficiency. The calculation formula is:

[0156]

[0157] in, represents the covariance, Represents variance.

[0158] 3) According to the correlation coefficient of each parameter , the reaction condition parameters are divided into three categories:

[0159] Key parameters: Parameters with higher correlation coefficients have a greater impact on reaction efficiency and are generally .

[0160] Secondary parameters: Parameters with medium correlation coefficients have a certain influence on the reaction efficiency. .

[0161] Slightly affected parameters: Parameters with lower correlation coefficients have less impact on reaction efficiency and are generally .

[0162] 4) Record the division results of these three types of parameters to provide a basis for subsequent model training and parameter optimization.

[0163] The role of this step is to conduct an in-depth analysis of the experimental data through the Bayesian optimization algorithm, determine the degree of influence of each reaction condition parameter on the reaction efficiency, and provide a basis for parameter classification for subsequent steps. Dividing parameters into key parameters, secondary parameters, and micro-influence parameters helps to better design model structure and optimization algorithm.

[0164] Step S30: Establishing a bidirectional neural network model for methyl oleate epoxidation reaction

[0165] The specific implementation process of this step is as follows:

[0166] 1) According to the three types of parameter features determined in step S20, a multi-branch tree-structured bidirectional neural network model is designed. The model includes the following sub-networks:

[0167] Key parameter sub-network: It uses a multi-layer perceptron network structure to process the impact of key parameters on reaction efficiency. The input is the key parameter vector , the output is the reaction efficiency index vector .

[0168] Secondary parameter subnetwork: A convolutional neural network structure is used to process the influence of secondary parameters on reaction efficiency. The input is the secondary parameter vector , the output is the reaction efficiency index vector .

[0169] Micro-influence parameter sub-network: It uses a recurrent neural network structure to process the influence of micro-influence parameters on reaction efficiency. The input is the micro-influence parameter vector , the output is the reaction efficiency index vector .

[0170] Overall evaluation sub-network: Using the Transformer network structure, using the self-attention mechanism and global information integration, comprehensively evaluate the impact of all parameters on reaction efficiency. The input is the complete reaction condition parameter vector , the output is the reaction efficiency index vector .

[0171] Fusion sub-network: Using the graph neural network structure, counterfactual reasoning mechanism and multimodal information fusion, the output of each sub-network is integrated to obtain the final reaction efficiency prediction. And the output of each sub-network, the output is the reaction efficiency index vector .

[0172] 2) Using the small test data collected in step S10 as the training set, train the established bidirectional neural network model. The input of the model is the reaction condition parameter vector , the output is the corresponding reaction efficiency index vector .

[0173] 3) During the training process, the back propagation algorithm is used to optimize the model parameters to minimize the predicted reaction efficiency The actual reaction efficiency The mean square error between is the target:

[0174]

[0175] in, is the number of training samples, are model parameters.

[0176] The purpose of this step is to establish a bidirectional neural network model covering different types of parameters, which can accurately predict the efficiency of methyl oleate epoxidation reaction. The multi-branch tree structure of the model can effectively capture the influence mechanism of different types of parameters on reaction efficiency and improve the prediction accuracy of the model.

[0177] Step S40: Generate a set of reaction condition parameters to be selected according to the reaction efficiency requirements

[0178] The specific implementation process of this step is as follows:

[0179] 1) Determine the range of target reaction efficiency, such as product yield, based on actual production needs Not less than 85%, selective Not less than 90%, conversion rate Not less than 95%.

[0180] 2) Appropriately discretize these reaction efficiency requirements, for example, divide the yield into three levels of 85%, 87%, and 90%, the selectivity into three levels of 90%, 92%, and 95%, and the conversion rate into three levels of 95%, 97%, and 99%. In this way, 27 (3x3x3) different reaction efficiency requirement combinations are obtained. .

[0181] 3) Combine each reaction efficiency requirement Input into the bidirectional neural network model established in step S30 to obtain the corresponding reaction condition parameter vector .

[0182] 4) All output reaction condition parameter vectors Combined into a set of candidate reaction condition parameters, laying the foundation for subsequent optimization.

[0183] The purpose of this step is to generate a set of candidate reaction condition parameters that meet the requirements of various reaction efficiency indicators based on actual production needs. Using the established bidirectional neural network model, parameter configuration solutions that meet different efficiency requirements can be quickly obtained, providing sufficient parameter candidate space for subsequent optimization.

[0184] Step S50: Optimizing the reaction condition parameter set to be selected using a genetic algorithm

[0185] The specific implementation process of this step is as follows:

[0186] 1) The reaction condition parameter set to be selected generated in step S40 As the initial population of the genetic algorithm.

[0187] 2) Design the fitness function of the genetic algorithm as the weighted sum of the reaction efficiency data, that is:

[0188]

[0189] in, is the fitness function value; , , They are product yield, selectivity and conversion rate; , , It is the weight coefficient set according to actual needs, satisfying .

[0190] 3) Set the crossover function of the genetic algorithm to arithmetic crossover and the mutation function to weighted non-uniform mutation. The weighted non-uniform mutation formula is:

[0191]

[0192] in, is the first parameter values; is the first parameter values; is the current iteration number; It is The upper limit of the parameters; It is The correlation coefficient of the parameters; is a function that controls the amplitude of variation and is defined as:

[0193]

[0194] in, is the maximum number of iterations; is a A random number between ; is the shape parameter, usually 2 or 3.

[0195] 4) Perform multiple rounds of genetic algorithm iterative optimization until the preset stop condition is met, and obtain the optimal reaction condition parameter vector that meets the target reaction efficiency requirements .

[0196] The purpose of this step is to use the genetic algorithm to iteratively optimize the candidate parameter set generated in step S40 to obtain the optimal reaction condition parameter combination that meets the target reaction efficiency requirements. The genetic algorithm is an effective global optimization algorithm that can quickly find a parameter configuration close to the optimal solution through fitness function evaluation and genetic operations. In particular, the design of weighted non-uniform mutation can better utilize the correlation information of parameters and improve the optimization efficiency.

[0197] Step S60: Screening intermediate reaction condition parameters based on weighted similarity

[0198] The specific implementation process of this step is as follows:

[0199] 1) Define weighted similarity index , used to evaluate intermediate reaction condition parameters Target reaction condition parameters Similarity between:

[0200]

[0201] in, is the weighted similarity; It is The correlation coefficient of the parameters; is the intermediate reaction condition parameter Values; is the target reaction condition parameter Values; and They are The upper and lower limits of the parameters; is the total number of parameters.

[0202] 2) In each iteration of the genetic algorithm, select the weighted similarity The intermediate reaction condition parameters that are greater than a preset similarity threshold (e.g., 0.9) are used as a screening set.

[0203] 3) The target reaction condition parameters obtained in step S50 are combined with the screening set Combine to form a seed reaction condition parameter set.

[0204] The role of this step is to select parameter combinations that are more similar to the target reaction condition parameters from the intermediate results of the genetic algorithm iteration through the weighted similarity index, and use them as the seed parameter set for the subsequent orthogonal test. This can minimize the scope of parameter optimization while ensuring the reaction efficiency, thereby improving the efficiency of the orthogonal test.

[0205] Step S70: Optimizing reaction condition parameters based on orthogonal experimental design

[0206] The specific implementation process of this step is as follows:

[0207] 1) Based on the seed reaction condition parameter set obtained in step S60, according to the correlation coefficient of each parameter As an important factor, an orthogonal test scheme is designed. Orthogonal test can fully explore the parameter space within a limited number of experiments and find the optimal parameter combination that is most similar to the target reaction efficiency. As an importance factor, we can better focus on the key parameters that have a greater impact on reaction efficiency.

[0208] 2) Carry out the experiment according to the orthogonal experimental scheme to obtain the experimental data set. Each data in the experimental data set contains the reaction condition parameter vector and the corresponding reaction efficiency index vector .

[0209] 3) In the experimental data set, calculate the reaction efficiency data of each set of experimental data and the preset target reaction efficiency data The similarity between :

[0210]

[0211] in, is the reaction efficiency data similarity; , , are the product yield, selectivity and conversion rate of the experimental data respectively; , , They are the product yield, selectivity and conversion of the target reaction efficiency data respectively; , , is the weight coefficient of the corresponding indicator. Select the similarity between the reaction efficiency data and the preset target reaction efficiency data The highest experimental data, the corresponding reaction condition parameter vector That is the optimal reaction condition parameter.

[0212] The purpose of this step is to further optimize and verify the seed parameter set obtained in step S60 by means of orthogonal experiments. Orthogonal experiments can fully explore the parameter space within a limited number of experiments and find the optimal parameter combination that is most similar to the target reaction efficiency. As an importance factor, we can better focus on the key parameters that have a greater impact on reaction efficiency.

[0213] Step S80: Determine the final optimal reaction condition parameters

[0214] The specific implementation process of this step is as follows:

[0215] 1) In the orthogonal test data set of step S70, find the reaction efficiency data Responding to the preset target efficiency data Similarity The highest experimental data.

[0216] 2) The reaction condition parameter vector corresponding to the experimental data Determine the optimal reaction condition parameters.

[0217] The purpose of this step is to determine the optimal reaction condition parameter combination that meets the target reaction efficiency requirements based on the optimization results of the previous steps. Through a series of analysis, modeling and optimization, the advantages of experimental data and model prediction are fully utilized to finally determine the reaction condition parameters that can best meet production needs.

[0218] In summary, this parameter determination method for improving the efficiency of methyl oleate epoxidation reaction fully utilizes the advantages of data-driven and model prediction, can greatly improve the efficiency of methyl oleate epoxidation reaction, and has strong practical value. The method mainly includes the following steps:

[0219] 1) Collect small-scale experimental data to obtain reaction condition parameters and reaction efficiency indicators.

[0220] 2) Use the Bayesian optimization algorithm to analyze the pilot test data and determine the correlation and importance of each parameter.

[0221] 3) Establish a bidirectional neural network model with a multi-branch tree structure to comprehensively capture the impact of parameters on reaction efficiency.

[0222] 4) Generate a set of candidate reaction condition parameters based on the target reaction efficiency requirements.

[0223] 5) Use genetic algorithm to optimize the set of candidate parameters and obtain the optimal parameters that meet the requirements.

[0224] 6) Use weighted similarity to filter intermediate results and determine the seed parameter set.

[0225] 7) Based on orthogonal experiments, the seed parameter set is further optimized to obtain the final optimal parameters.

[0226] The method of Example 1 fully combines technologies such as experimental data analysis, machine learning modeling and optimization algorithms, systematically solves the problem of improving the efficiency of methyl oleate epoxidation reaction, and provides a feasible solution for related production practices.

[0227] In order to further better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: A chemical company recently carried out research on the epoxidation reaction of methyl oleate, hoping to improve the product yield and selectivity by optimizing the reaction conditions to meet the needs of downstream customers. After preliminary small-scale experiments, 20 groups of experimental data under different reaction conditions were collected, as shown in Table 1.

[0228] Table 1. Experimental data of methyl oleate epoxidation reaction

[0229] Experiment number Reaction temperature Reaction time Catalyst dosage Oxidant dosage Stirring speed Reaction pressure Raw material concentration pH Solvent type Solvent dosage Reactor Type Reaction material ratio Additive Type Additive dosage Product yield Selectivity Conversion rate 1 60 °C 2 h 0.5wt% 1.2eq. 500rpm 1 atm 0.2mol / L 6.5 Ethanol 10 mL Reflux Reactor 1:1 - - 82% 88% 90% 2 65 °C 2.5 h 0.6wt% 1.4eq. 550rpm 1 atm 0.25mol / L 6.8 Ethanol 12 mL Reflux Reactor 1:1 - - 84% 90% 92% ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... 20 70 °C 3 h 0.8wt% 1.6eq. 600rpm 1 atm 0.3mol / L 7.2 Ethanol 15 mL Reflux Reactor 1:1 Sulfamic acid 0.1wt% 88% 92% 95%

[0230] Based on these small-scale experimental data, the chemical company adopted the method proposed in the present invention to optimize the parameters of the epoxidation reaction of methyl oleate, and the specific steps are as follows:

[0231] Step S10: Collecting small test data

[0232] The chemical company first sorted out the main reaction condition parameters that affect the efficiency of methyl oleate epoxidation reaction, including 14 parameters such as reaction temperature, reaction time, catalyst dosage, oxidant dosage, stirring speed, reaction pressure, raw material concentration, pH value, solvent type, solvent dosage, reactor type, reaction material ratio, additive type and additive dosage. Subsequently, 20 groups of small-scale experiments were designed and carried out to collect reaction efficiency data under the above parameter combinations, including product yield, selectivity and conversion rate. The experimental data are shown in Table 1.

[0233] Step S20: Parameter importance analysis

[0234] The chemical company used the Bayesian optimization algorithm to analyze the pilot test data and obtained the correlation coefficient between each reaction condition parameter and the reaction efficiency index, as shown in Table 2.

[0235] Table 2. Correlation coefficients between reaction condition parameters and reaction efficiency indicators

[0236] parameter Yield correlation Selective relevance Conversion rate correlation Reaction temperature 0.78 0.72 0.65 Reaction time 0.72 0.68 0.61 Catalyst dosage 0.75 0.70 0.63 Oxidant dosage 0.68 0.64 0.58 Stirring speed 0.52 0.48 0.45 Reaction pressure 0.39 0.35 0.32 Raw material concentration 0.63 0.59 0.54 pH 0.46 0.42 0.40 Solvent type 0.31 0.29 0.27 Solvent dosage 0.42 0.38 0.35 Reactor Type 0.29 0.27 0.25 Reaction material ratio 0.35 0.33 0.30 Additive Type 0.23 0.22 0.21 Additive dosage 0.27 0.25 0.23

[0237] According to the data in Table 2, chemical companies divide these parameters into three categories:

[0238] Key parameters: reaction temperature (0.78), reaction time (0.72), catalyst dosage (0.75)

[0239] Secondary parameters: oxidant dosage (0.68), raw material concentration (0.63)

[0240] Slightly affected parameters: stirring speed (0.52), pH value (0.46), solvent dosage (0.42)

[0241] Step S30: Establish a bidirectional neural network model

[0242] Based on the parameter importance analysis results obtained in step S20, the chemical company designed a bidirectional neural network model with a multi-branch tree structure, including the following sub-networks:

[0243] Key parameter sub-network: It adopts a multi-layer perceptron network structure, with key parameters such as reaction temperature, reaction time and catalyst dosage as input, and reaction efficiency indicators as output.

[0244] Secondary parameter subnetwork: It adopts a convolutional neural network structure, with secondary parameters such as oxidant dosage and raw material concentration as input, and the output is the reaction efficiency index.

[0245] Micro-influence parameter sub-network: It adopts a recurrent neural network structure, with micro-influence parameters such as stirring speed, pH value and solvent dosage as input, and the output is the reaction efficiency index.

[0246] Overall evaluation sub-network: adopts the Transformer network structure, comprehensively considers the impact of all parameters on reaction efficiency, and outputs the reaction efficiency index.

[0247] Fusion sub-network: Using graph neural network structure, the outputs of each sub-network are fused to give the final reaction efficiency prediction.

[0248] The bidirectional neural network model uses the pilot test data in Table 1 as the training set, and uses the back propagation algorithm to optimize the model parameters to minimize the error between the predicted reaction efficiency and the actual reaction efficiency. After training, the average absolute error of the model on the validation set is 2.8%.

[0249] Step S40: Generate a set of parameters to be selected

[0250] In response to the needs of downstream customers, the chemical company determined the target reaction efficiency indicators: product yield is not less than 90%, selectivity is not less than 92%, and conversion rate is not less than 95%. These target indicators are discretized to obtain 27 (3×3×3) different reaction efficiency demand combinations, as shown in Table 3.

[0251] Table 3. Reaction efficiency requirement combination

[0252] Serial number Product yield Selectivity Conversion rate 1 90% 92% 95% 2 90% 92% 97% 3 90% 92% 99% 4 90% 95% 95% ... ... ... ... 27 92% 95% 99%

[0253] By inputting these 27 reaction efficiency requirement combinations into the bidirectional neural network model established in step S30, a reaction condition parameter combination that meets each requirement can be quickly obtained to form a candidate parameter set.

[0254] Step S50: Genetic Algorithm Optimization

[0255] The chemical company uses the candidate parameter set obtained in step S40 as the initial population of the genetic algorithm. The fitness function of the genetic algorithm is designed to be the weighted sum of the reaction efficiency index:

[0256]

[0257] in, is the fitness function value; is the product yield; is selective; It's the conversion rate.

[0258] The crossover function of the genetic algorithm uses arithmetic crossover, and the mutation function uses weighted non-uniform mutation:

[0259]

[0260] in, is the first parameter values; is the first parameter values; is the current iteration number; It is The upper limit of the parameters; It is The correlation coefficient of the parameters; The definition of is the same as above.

[0261] After 50 rounds of iterative optimization, the genetic algorithm obtained the optimal reaction condition parameter combination that met the target reaction efficiency requirements (yield ≥ 90%, selectivity ≥ 92%, conversion rate ≥ 95%): reaction temperature 68°C, reaction time 2.8 h, catalyst dosage 0.65 wt%, oxidant dosage 1.5 eq., stirring speed 580 rpm, reaction pressure 1 atm, raw material concentration 0.26 mol / L, pH 7.0, ethanol as solvent, solvent dosage 13 mL, reflux reactor, reaction material ratio 1:1, and addition of 0.08 wt% aminosulfonic acid.

[0262] Step S60: Filtering intermediate results

[0263] In each iteration of the genetic algorithm, the chemical company calculates the weighted similarity between all intermediate parameter combinations and the optimal parameter combination. :

[0264]

[0265] in, It is The correlation coefficient of the parameters is determined by step S20; is the first combination of intermediate parameters Values; is the optimal parameter combination Values; and They are The upper and lower limits of a parameter.

[0266] Chemical companies will weight similarity The intermediate parameter combinations greater than 0.9 are used as the screening set and merged with the optimal parameter combination obtained by the genetic algorithm to form the seed parameter set.

[0267] Step S70: Orthogonal test verification

[0268] Based on the seed parameter set obtained in step S60, the chemical company designed a set of orthogonal experiments and set the importance coefficient of each parameter As the importance factor. The experimental results are shown in Table 4.

[0269] Table 4. Orthogonal test results

[0270] Experiment number Reaction temperature Reaction time Catalyst dosage Oxidant dosage Stirring speed Reaction pressure Raw material concentration pH Solvent type Solvent dosage Reactor Type Reaction material ratio Additive Type Additive dosage Product yield Selectivity Conversion rate 1 65 °C 2.5 h 0.6wt% 1.4eq. 560rpm 1 atm 0.24mol / L 6.8 Ethanol 12 mL Reflux Reactor 1:1 - - 87% 91% 94% 2 67 °C 2.7 h 0.64wt% 1.5eq. 570rpm 1 atm 0.25mol / L 6.9 Ethanol 13 mL Reflux Reactor 1:1 Sulfamic acid 0.07wt% 89% 92% 95% 3 68 °C 2.8 h 0.65wt% 1.5eq. 580rpm 1 atm 0.26mol / L 7.0 Ethanol 13 mL Reflux Reactor 1:1 Sulfamic acid 0.08wt% 90% 93% 96% 4 69 °C 2.9 h 0.67wt% 1.6eq. 590rpm 1 atm 0.27mol / L 7.1 Ethanol 14 mL Reflux Reactor 1:1 Sulfamic acid 0.09wt% 91% 93% 97%

[0271] In the orthogonal test results, the reaction efficiency data corresponding to Experiment 3 (yield 90%, selectivity 93%, conversion rate 96%) are closest to the target reaction efficiency requirements. Therefore, the chemical company determined the reaction condition parameters of Experiment 3 as the final optimal parameter combination.

[0272] Step S80: Determine the optimal parameters

[0273] According to the orthogonal test results, the chemical company set the reaction temperature of 68°C, reaction time of 2.8 h, catalyst dosage of 0.65wt%, oxidant dosage of 1.5 eq., stirring speed of 580 rpm, reaction pressure of 1 atm, raw material concentration of 0.26 mol / L, pH value of 7.0, ethanol as solvent, solvent dosage of 13 mL, reflux reactor, reaction material ratio of 1:1, and addition of 0.08 wt% aminosulfonic acid as the final optimal reaction condition parameters. This set of parameters meets the target reaction efficiency requirements determined by the company: product yield is not less than 90%, selectivity is not less than 92%, and conversion rate is not less than 95%.

[0274] Finally, the chemical company conducted a scaled-up experiment based on this set of optimal reaction condition parameters. The results showed that in a 100 L scale reactor, the product yield reached 91%, the selectivity was 93%, and the conversion rate was 96%, which was basically consistent with the results of the small-scale experiment, verifying the effectiveness of the method of the present invention.

[0275] Through the above steps, the chemical company successfully applied the parameter determination method for improving the efficiency of methyl oleate epoxidation reaction proposed by the present invention, systematically analyzed and optimized the reaction condition parameters, and finally determined the optimal parameter combination that meets production needs. The main advantages of this method are reflected in the following aspects:

[0276] 1. The parameter importance analysis of the Bayesian optimization algorithm provides a basis for subsequent model design and optimization, avoiding the neglect of parameter coupling relationships in traditional single-factor experimental methods or orthogonal experimental methods.

[0277] 2. The multi-branch tree structure of the bidirectional neural network model can effectively capture the influence mechanism of different types of parameters on reaction efficiency and improve the prediction accuracy of the model.

[0278] 3. The weighted non-uniform mutation design of the genetic algorithm makes full use of the parameter importance information and improves the convergence of the algorithm and the quality of the solution.

[0279] 4. Weighted similarity screening and orthogonal test verification narrow the search space for parameter optimization and improve the reliability of the final solution.

[0280] Through the organic combination of the above technical means, the method of the present invention can quickly and effectively determine the optimal reaction condition parameters that meet the target reaction efficiency requirements, greatly improve the efficiency of the methyl oleate epoxidation reaction, and provide a feasible solution for related production practices. In addition, the method also has good versatility and can be extended to other similar chemical reaction process optimization.

[0281] In summary, the chemical company successfully optimized the reaction conditions of the epoxidation reaction of methyl oleate by adopting the parameter determination method proposed in the present invention, and the product yield, selectivity and conversion rate all reached the expected goals, which fully verified the practical value of the method. In the future, the company also plans to further expand the production scale and apply this method to the optimization of other chemical reaction processes of the company to continuously improve product competitiveness and meet the needs of downstream customers.

[0282] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction, characterized in that: The following steps are involved: S10, collecting multiple groups of small-scale test data of methyl oleate epoxidation reaction, including reaction condition parameters and reaction efficiency data; S20, analyzing the pilot test data using a Bayesian optimization algorithm to obtain a correlation coefficient between each reaction condition parameter and the reaction efficiency data, and clustering the correlation coefficients to obtain three types of reaction condition parameters, specifically key parameters, secondary parameters, and micro-influence parameters; S30, establishing a training data set according to the multiple groups of small test data, training a bidirectional neural network model with a multi-branch tree structure, and obtaining a bidirectional model for the epoxidation reaction of methyl oleate; wherein the input of the training is the reaction condition parameter, and the output of the training is the reaction efficiency data; S40, discretizing the preset reaction efficiency requirement range to obtain multiple reaction efficiency requirements, inputting each reaction efficiency requirement into the bidirectional model of the methyl oleate epoxidation reaction to obtain corresponding reaction condition parameters, and merging the reaction condition parameters corresponding to each reaction efficiency requirement into a set of reaction condition parameters to be selected; S50, using a genetic algorithm, taking the set of reaction condition parameters to be selected as the initial population, taking the reaction efficiency data as the fitness function, setting the crossover function as arithmetic crossover, and the mutation function as weighted non-uniform mutation, performing iterative optimization, and finally obtaining the target reaction condition parameters that meet the iteration stop condition; S60, among the intermediate reaction condition parameters outputted in each iteration of the genetic algorithm, selecting the intermediate reaction condition parameters whose weighted similarity with the target reaction condition parameters is greater than the weighted similarity threshold as a screening set, and merging the screening set and the target reaction condition parameters into a seed reaction condition parameter set; S70, designing an orthogonal experiment according to the seed reaction condition parameter set with the correlation coefficient of each reaction condition parameter as an importance factor, and obtaining an experimental data set of the orthogonal experiment, wherein each experimental data in the experimental data set includes reaction condition parameters and reaction efficiency data; S80, selecting, from the experimental data set, reaction condition parameters of experimental data whose reaction efficiency data has the highest similarity to preset target reaction efficiency data as optimal reaction condition parameters; The bidirectional neural network model includes a key parameter subnetwork, a secondary parameter subnetwork, a micro-influence parameter subnetwork, an overall evaluation subnetwork and a fusion subnetwork. The tree structure is as follows: the outputs of the key parameter subnetwork, the secondary parameter subnetwork, the micro-influence parameter subnetwork and the overall evaluation subnetwork are the inputs of the fusion subnetwork. The key parameter sub-network is used to process key parameters, the training input is the key parameters, the training output is the reaction efficiency data, and its structure is a multi-layer perceptron network; The secondary parameter subnetwork is used to process secondary parameters, the training input is the secondary parameters, the training output is the reaction efficiency data, and its structure is a convolutional neural network; The micro-influence parameter sub-network is used to process micro-influence parameters, the training input is the micro-influence parameters, the training output is the reaction efficiency data, and its structure is a recurrent neural network; The key parameter is a parameter with a correlation coefficient greater than 0.7; the secondary parameter is a parameter with a correlation coefficient greater than or equal to 0.4 and less than or equal to 0.7; and the micro-influence parameter is a parameter with a correlation coefficient less than 0.

4.

2. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 1, characterized in that: The reaction efficiency data include product yield, selectivity and conversion rate.

3. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 1, characterized in that: The reaction condition parameters include reaction temperature, reaction time, catalyst dosage, oxidant dosage, stirring speed, reaction pressure, raw material concentration, pH value, solvent type, solvent dosage, reactor type, reaction material ratio, additive type, and additive dosage.

4. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 1, characterized in that: The weight of the weighted similarity is the correlation coefficient corresponding to each reaction condition parameter.

5. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 1, characterized in that: The variation function is a weighted non-uniform variation, specifically, a weighted non-uniform variation is performed using the correlation coefficient of the reaction condition parameters as a weight.

6. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 1, characterized in that: The overall evaluation sub-network is used to adopt a self-attention mechanism and global information integration. The training input is all reaction condition parameters, and the training output is reaction efficiency data. Its structure is a Transformer network.

7. The method for determining parameters for improving the efficiency of methyl oleate epoxidation reaction according to claim 6, characterized in that: The fusion sub-network is used to adopt a counterfactual reasoning mechanism and multimodal information fusion. The training input is all reaction condition parameters and the output of each sub-network. The training output is reaction efficiency data. Its structure is a graph neural network.

Images