LDB product reaction optimization control method based on data mining

Abstract

The invention discloses an LDB product reaction optimization control method based on data mining, and the method comprises the steps: collecting and arranging an experiment condition of an LDB sample as a modeling feature, collecting and arranging an LDB experiment reaction result as a modeling target, and constructing a modeling data set based on the modeling feature and the modeling target; carrying out modeling feature preprocessing on the data set, distinguishing superior reactions and inferior reactions according to reaction yield, and respectively calculating single-class samples of the superior reactions and inferior reactions to obtain a combined feature vector; and drawing a scatter diagram based on the combined feature vector, obtaining an area in which the proportion of the optimal reaction exceeds a preset threshold value, and carrying out statistics to obtain a modeling feature scheme of the optimal reaction. According to the method, dozens of LDB experiment conditions can be treated at the same time, the change trend of superior reactions and inferior reactions is analyzed, an experiment condition optimization scheme is provided for experimenters, and the reaction yield of LDB products is effectively increased.

Description

Technical Field

[0001] The present invention belongs to the technical field of pharmaceutical intermediate preparation, and in particular relates to an LDB product reaction optimization control method based on data mining. Background Art

[0002] LDB, or 3-bromo-1-(3-chloro-2-pyridyl)-1H-pyrazole-5-carboxylic acid, is an important intermediate of chlorantraniliprole, a new class of highly effective and low-toxic o-formylaminobenzamide insecticides that plays a vital role in agricultural pest control. Therefore, efficient and stable LDB production is crucial to ensuring the supply of high-quality pesticides to downstream manufacturers.

[0003] Currently, the synthesis of LDB products relies heavily on manual experimentation to optimize reaction yields. A complete LDB reaction involves two to four chemical steps, dozens of experimental procedures, and 30 to 40 conditions that require precise control. Within a limited experimental cycle, researchers must combine and adjust numerous conditions to find the optimal solution. This process requires significant manpower and is time-consuming, with lengthy experimental cycles and high costs.

[0004] Due to the complexity of the LDB reaction and the multitude of experimental conditions, it is difficult to explore and optimize the process through manual experience and trial and error. On the one hand, it is difficult for experimenters to fully cover all possible combinations of experimental conditions within a limited number of experiments, making it difficult to accurately find the optimal experimental condition scheme. On the other hand, faced with a large amount of complex experimental conditions and reaction result data, it is impossible to accurately grasp the changing trends of superior and inferior reactions, and thus it is impossible to provide targeted experimental condition optimization directions for subsequent experiments. These defects seriously restrict the improvement of the reaction yield of LDB products and limit the further improvement of production efficiency and economic benefits. Summary of the Invention

[0005] To address the problems of the prior art, the present invention provides a data-mining-based method for optimizing and controlling the reaction yield of LDB products, aiming to design LDB products with high reaction yields. Leveraging machine learning algorithms, this method can simultaneously process dozens of LDB experimental conditions. Through data integration and algorithmic calculations, it analyzes the changing trends of superior and inferior reactions under different experimental conditions and summarizes the modeling feature ranges of superior reactions. Based on this analysis, it provides experimenters with a scientific and reasonable solution for optimizing experimental conditions, effectively reducing the number of experiments and experimental cycles, lowering labor and time costs, and effectively improving the reaction yield of LDB products.

[0006] To achieve the above objectives, the present invention provides a method for optimizing the reaction of LDB products based on data mining, comprising:

[0007] Collecting and organizing the experimental conditions of the LDB samples as modeling features, collecting and organizing the LDB experimental reaction results as modeling targets, and constructing a modeling dataset based on the modeling features and modeling targets;

[0008] After preprocessing the data set for modeling features, distinguishing superior reactions from inferior reactions based on reaction yields, and calculating single-class samples of the superior reactions and inferior reactions respectively to obtain a combined feature vector;

[0009] A scatter plot is drawn based on the combined feature vector to obtain the area where the proportion of the excellent class reaction exceeds a preset threshold, and a modeling feature scheme of the excellent class reaction is obtained by statistics.

[0010] Preferably, the modeling objectives include reaction yield and product content.

[0011] Preferably, the process of preprocessing the data set for modeling features includes:

[0012] Based on experimental experience, modeling features that have no effect on reaction yield were deleted;

[0013] Delete low-variance modeling features and strongly correlated modeling features; wherein, the low-variance modeling features are modeling features with variance < 0.0001, and the strongly correlated modeling features are modeling features with correlation > 0.95.

[0014] Preferably, the process of distinguishing superior reactions from inferior reactions based on reaction yield comprises:

[0015] Reactions with a reaction yield greater than or equal to the mean are classified as superior reactions;

[0016] Reactions with reaction yields less than the mean are classified as inferior reactions.

[0017] Preferably, the process of calculating the single-class samples of the excellent and inferior reactions respectively includes:

[0018] Obtain the mean of the modeling features of the single-class samples, subtract the mean of the modeling features of the single-class samples from the modeling feature values ​​of the single-class samples, calculate the Gram matrix, and obtain the discrete matrix of the single-class samples;

[0019] Adding the discrete matrices of the two single-class samples to obtain a cumulative discrete matrix, and calculating the inverse matrix of the cumulative discrete matrix;

[0020] Calculate the difference matrix of two single-class samples and calculate the combined characteristic coefficient vector;

[0021] Based on the combined feature coefficient vector, a combined feature vector is calculated.

[0022] Preferably, the discrete matrix D of the single-class samples label The formula expression is:

[0023]

[0024] Among them, X label is the modeling feature value of a single class sample, is the modeling feature mean of a single class of samples.

[0025] Preferably, the formula expression of the cumulative discrete matrix is:

[0026]

[0027] The formula expression of the inverse matrix of the cumulative discrete matrix is:

[0028]

[0029] Among them, D label is the discrete matrix of single-class samples.

[0030] Preferably, the formula for calculating the difference matrix of two single-class samples is:

[0031]

[0032] in, represents the mean of the modeling features of the superior response, The mean of the modeled features representing the inferior response;

[0033] The formula expression of the combined characteristic coefficient vector is:

[0034] V fis =D diff SD -1

[0035] Among them, D diff Represents the difference matrix of two single-class samples, SD -1 Preferably, based on the combined characteristic coefficient vector, the formula for calculating the combined characteristic vector includes:

[0036] C1=XV fis ,C2=XV sec

[0037] Among them, C1 and C2 represent the combined feature vectors, V fis and V sec represents the combined characteristic coefficient vector.

[0038] Preferably, the modeling feature scheme for statistically obtaining the superior response includes:

[0039] The minimum, maximum, and mean values ​​of the modeling features of the statistically superior response.

[0040] Compared with the prior art, the present invention has the following advantages and technical effects:

[0041] The present invention provides a method for optimizing and controlling the reaction yield of LDB products based on data mining. The method can simultaneously process dozens of LDB experimental conditions, analyze the changing trends of superior and inferior reactions, and provide experimental personnel with an experimental condition optimization scheme, thereby effectively achieving the purpose of improving the reaction yield of LDB products.

[0042] The present invention proposes a method for optimizing the reaction yield of LDB products based on data mining. By integrating data and performing algorithmic calculations, the distribution patterns of superior and inferior reactions are analyzed, and the modeling feature range of superior reactions is summarized. In practical applications, this method can provide data support for the optimization of LDB reactions, thereby effectively improving the LDB reaction yield, enhancing product quality, and enhancing market competitiveness.

[0043] In summary, machine learning methods based on data mining can provide an efficient and accurate solution for optimizing LDB product reactions. Therefore, the data mining-based method for optimizing and controlling the reaction yield of LDB products proposed in this paper has important practical application value and broad market prospects, and is expected to promote the optimization and upgrading of LDB product synthesis processes and the sustainable development of the chlorantraniliprole industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0045] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of sample distribution trend in the principal component space according to an embodiment of the present invention;

[0047] Figure 3 Schematic diagram of LDB sample distribution trend in the C1×C2 space according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0049] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0050] Example 1

[0051] like Figure 1-3 As shown, this embodiment provides an LDB product reaction optimization control method based on data mining, including:

[0052] S1. Collect and organize the experimental conditions of LDB samples as modeling features, collect and organize the LDB experimental reaction results as modeling targets, and construct a modeling dataset based on the modeling features and modeling targets;

[0053] S2. performing modeling feature preprocessing on the data set to reduce redundant information of the modeling features;

[0054] S3. Distinguish superior reactions from inferior reactions based on reaction yields. For superior and inferior reactions, perform the following calculations on the single-class samples to obtain the combined feature vector:

[0055] 1) Calculate the mean of the modeling features of a single class of samples, recorded as in, Represents the mean, label represents a class in the good or bad class, such as The mean of the modeled features representing the superior response;

[0056] 2) Modeling feature value of a single class of samples, denoted as X label ; First, model the eigenvalue X of the single-class sample label Subtract the modeling feature mean of a single class sample Right now Then calculate the Gram matrix to obtain the discrete matrix D of the single class sample label ,Right now

[0057] 3) Add the discrete matrices of the two single-class samples to obtain the cumulative discrete matrix SD = ∑D label ; Then calculate the inverse matrix SD of the cumulative discrete matrix -1 =(∑D label ) -1 .

[0058] 4) Calculate the difference matrix of two single-class samples Calculate the combined characteristic coefficient vector V fis =D diff SD -1 .

[0059] 5) Calculate the dispersion coefficient matrix S:

[0060]

[0061] 6) Calculate the combined characteristic coefficient vector V sec :

[0062] V sec =D diff [SD -1 -S(SD -1 SD -1 )]

[0063] 7) Based on the combined characteristic coefficient vector V fis and V sec , calculate the combined eigenvectors C1 and C2:

[0064] C1=XV fis ,C2=XV sec

[0065] S4. Draw a scatter plot based on the combined eigenvectors C1 and C2. The sample points in the plot are marked with different shapes to distinguish between superior and inferior reactions. Identify areas in the scatter plot with a high proportion of superior reactions and calculate the modeling features of the superior reactions to guide the next step of the experiment and improve the reaction yield.

[0066] Furthermore, the S1 collects and organizes the LDB experimental reaction results as modeling targets, including reaction yield, product content, etc.;

[0067] Furthermore, the method for performing modeling feature preprocessing in S2 is: based on experimental experience, deleting modeling features that have almost no effect on the reaction yield;

[0068] Furthermore, the method for S2 to perform modeling feature preprocessing is: deleting low-variance modeling features and strongly correlated modeling features; wherein, the low-variance modeling features are modeling features with variance <0.0001, and the strongly correlated modeling features are modeling features with correlation >0.95.

[0069] Furthermore, the S3 distinguishes superior reactions from inferior reactions according to the reaction yield, including: a reaction yield greater than or equal to the average is an superior reaction, and a reaction yield less than the average is an inferior reaction.

[0070] Furthermore, the S4 statistical modeling characteristics of the superior reaction include: minimum, maximum, mean and other statistical data of the modeling characteristics that are of practical significance to the experimental reaction.

[0071] Example 2

[0072] This embodiment provides a method for optimizing and controlling the reaction yield of an LDB product based on data mining, comprising:

[0073] S1. Organize the reaction conditions and yields of 273 LDB samples and construct a dataset based on these experimental conditions and yields. For each LDB sample, 38 reaction condition variables were organized as modeling features. Each LDB sample was assigned a reaction yield as a target, creating a dataset that included both reaction condition features and yields. The mean reaction yield in this dataset was 74.45%.

[0074] S2. Perform modeling feature preprocessing on the dataset described in S1. Based on experimental experience, first remove modeling features that have very little impact on reaction yield; then remove low-variance and strongly correlated modeling features. This can reduce the number of modeling features from 38 to 29.

[0075] S3. Define superior reactions and inferior reactions based on the mean reaction yield, where the reaction yield is greater than or equal to the mean and the reaction yield is less than the mean and the reaction is inferior.

[0076] Calculate the characteristic mean of the superior response The characteristic mean of the superior response in Represents the mean. Then calculate the discrete matrix D of the optimal response 优类 , the discrete matrix D of the optimal response 劣类 , cumulative discrete matrix SD, the calculation formula is as follows:

[0077]

[0078] SD=D 优类 +D 劣类

[0079] Calculate the difference matrix D between the best and the best responses in turn diff , combined characteristic coefficient vector V fis , combined eigenvector C1, the calculation formula is as follows:

[0080]

[0081] V fis =D diff SD -1

[0082] C1=XV fis

[0083] Calculate the discrete coefficient matrix S and the combined characteristic coefficient vector V in turn sec , combined eigenvector C2, the calculation formula is as follows:

[0084]

[0085] S4. Based on the steps described in S3, the combined feature vectors C1 and C2 are obtained. A two-dimensional scatter plot of the LDB data is constructed based on C1 and C2. The scatter plot can reflect the distribution trend of the LDB samples in the C1×C2 space.

[0086] For example, for the LDB data sets S1 and S2, the principal component method is used to calculate the principal component 1 and the principal component 2 based on the modeling features, and a two-dimensional scatter plot is constructed based on the principal component 1 and the principal component 2. The results are as follows: Figure 2 As shown, dots represent LDB-excellent reactions, and crosses represent LDB-excellent reactions. Figure 2 As can be seen in the figure, the distribution areas of excellent and inferior responses completely overlap, and it is impossible to distinguish obvious excellent samples or patterns of excellent responses. This shows that the conventional principal component method cannot meet the processing requirements of the LDB dataset.

[0087] For example, for the LDB data sets described in S1 and S2, the steps described in S3 are used to calculate the combined feature vectors C1 and C2 based on the modeling features. Based on C1 and C2, a two-dimensional scatter plot of the LDB data can be constructed. The result is as follows: Figure 3 As shown, dots represent LDB-excellent reactions, and crosses represent LDB-excellent reactions. Figure 3 As can be seen in the figure, the excellent reactions are mainly located in the middle and lower left parts, showing a trend from the middle to the lower left. The excellent reactions are mainly located in the middle and upper right parts, showing a trend from the middle to the upper right. Although there is a large overlap between the excellent reaction distribution and the excellent reaction distribution, the area with a large proportion of excellent reactions can be found in the upper right corner ( Figure 3 Therefore, the modeling result of the method of the present invention is better than the traditional modeling result.

[0088] Example 3

[0089] by Figure 3 The region with a high proportion of excellent reactions in the dataset was selected as the optimization zone, and the boundaries of this region were used as the optimization boundaries to propose optimization suggestions for the LDB reaction experimental conditions. The optimization suggestions are shown in Table 1, which lists the six experimental conditions with the largest variation in the optimization range. The remaining experimental conditions were treated as the mean of the training samples. Following the optimization suggestions in Table 1, a new LDB experiment was implemented, and the reaction yield increased from 74.45% in the dataset to 75.21%, an increase of 0.76%. Table 1 shows the optimization suggestions for the LDB reaction experimental conditions.

[0090] Table 1

[0091]

[0092]

[0093] Example 4

[0094] This embodiment further discloses a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first embodiment.

[0095] Example 5

[0096] This embodiment further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0097] Example 6

[0098] This embodiment further discloses a computer program product, including a computer program, which implements the steps of the method described in the first embodiment when executed by a processor.

[0099] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A LDB product reaction optimization control method based on data mining, characterized in that: include: Collecting and organizing the experimental conditions of the LDB samples as modeling features, collecting and organizing the LDB experimental reaction results as modeling targets, and constructing a modeling dataset based on the modeling features and modeling targets; After preprocessing the data set for modeling features, distinguishing superior reactions from inferior reactions based on reaction yields, and calculating single-class samples of the superior reactions and inferior reactions respectively to obtain a combined feature vector; A scatter plot is drawn based on the combined feature vector to obtain the area where the proportion of the excellent class reaction exceeds a preset threshold, and a modeling feature scheme of the excellent class reaction is obtained by statistics.

2. The method according to claim 1, characterized in that The modeling objectives include reaction yield and product content.

3. The method according to claim 1, characterized in that The process of preprocessing the dataset for modeling features includes: Based on experimental experience, modeling features that have no effect on reaction yield were deleted; Delete low-variance modeling features and strongly correlated modeling features; wherein, the low-variance modeling features are modeling features with variance < 0.0001, and the strongly correlated modeling features are modeling features with correlation > 0.

95.

4. The method according to claim 1, wherein The process of distinguishing superior reactions from inferior reactions based on reaction yield includes: Reactions with a reaction yield greater than or equal to the mean are classified as superior reactions; Reactions with reaction yields less than the mean are classified as inferior reactions.

5. The method according to claim 1, wherein The process of calculating the single-class samples of the excellent and inferior reactions respectively includes: Obtain the mean of the modeling features of the single-class samples, subtract the mean of the modeling features of the single-class samples from the modeling feature values ​​of the single-class samples, calculate the Gram matrix, and obtain the discrete matrix of the single-class samples; Adding the discrete matrices of the two single-class samples to obtain a cumulative discrete matrix, and calculating the inverse matrix of the cumulative discrete matrix; Calculate the difference matrix of two single-class samples and calculate the combined characteristic coefficient vector; Based on the combined feature coefficient vector, a combined feature vector is calculated.

6. The method according to claim 5, characterized in that The discrete matrix D of the single-class samples label The formula expression is: Among them, X label is the modeling feature value of a single class sample, is the mean of the modeling features of a single class of samples.

7. The method according to claim 5, characterized in that The formula expression of the cumulative discrete matrix is: SD=∑D label The formula expression of the inverse matrix of the cumulative discrete matrix is: SD -1 =(∑D label ) -1 Among them, D label is the discrete matrix of single-class samples.

8. The method according to claim 5, characterized in that The formula for calculating the difference matrix of two single-class samples is: in, represents the mean of the modeling features of the superior response, The mean of the modeled features representing the inferior response; The formula expression of the combined characteristic coefficient vector is: V fis =D diff SD -1 Among them, D diff Represents the difference matrix of two single-class samples, SD -1 Represents the inverse of the cumulative discrete matrix.

9. The method according to claim 5, characterized in that Based on the combined characteristic coefficient vector, the formula expression for calculating the combined characteristic vector includes: C1=XV fis ,C2=XV sec Among them, C1 and C2 represent the combined feature vectors, V fis and V sec represents the combined characteristic coefficient vector.

10. The method according to claim 1, characterized in that Modeling feature schemes for statistically obtaining superior responses include: The minimum, maximum, and mean values ​​of the modeling features of the statistically superior response.

Images