CSA cement retarder screening method based on machine learning and quantum chemistry

Abstract

The invention discloses a CSA cement retarder screening method based on machine learning and quantum chemistry. The CSA cement retarder screening method comprises the steps that a data set containing CSA cement retarders and corresponding molecular structures is obtained; establishing a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the multiple molecular descriptors comprise self-defined molecular composition descriptors, Morgan fingerprint descriptors and quantum chemistry descriptors; establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time; based on the quantitative structure-activity relationship model, calculating molecular quantitative performance parameters of different retarders on the CSA cement surface by adopting a quantum chemistry calculation method; and screening and outputting the CSA cement retarders meeting conditions according to the calculated molecular quantification performance parameters of the different retarders on the surface of the CSA cement. The cement retarder screening device improves the cement retarder screening accuracy and efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of civil engineering materials, and in particular to a CSA cement retarder screening method based on machine learning and quantum chemistry. Background Art

[0002] Calcium sulfoaluminate cement is considered to be a green, low-carbon and environmentally friendly cement due to its low carbon emissions and low energy consumption. Calcium sulfoaluminate (CSA) cement is often used in high temperature environments or projects that require rapid hardening. In actual projects, there are problems such as too fast setting and difficulty in pumping. Retarder admixtures are usually required to prevent premature solidification after mixing. The types of retarders commonly used currently mainly include inorganic and organic retarders, which achieve the retarding effect by inhibiting the cement hydration reaction rate and delaying the crystallization and setting process. The traditional screening method of retarders mainly uses experimental means to detect the setting time, fluidity, early strength and other indicators of different additives in cement. The limitations of these screening methods are mainly reflected in: Lack of a detailed understanding of the reaction mechanism: Traditional methods focus on macroscopic effects (such as setting time, strength, etc.), but lack in-depth research on the mechanism of action of retarders at the molecular and microscopic levels, resulting in the difficulty of achieving precise control of the optimization of retarders.

[0003] Due to the diversity of the molecular structure and function of retarders, the traditional method that relies on trial and error consumes a lot of manpower and time costs in finding suitable types and ratios of reactants. It is also difficult to accurately control the performance of retarders and cannot be efficiently customized to meet specific engineering needs.

[0004] Therefore, the prior art needs to be improved. Summary of the invention

[0005] The technical problem to be solved by the present invention is that, in view of the defects of the prior art, the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry to solve the problem that the traditional CSA cement retarder screening method is difficult to accurately control the retarder performance, and thus cannot be efficiently customized to meet specific engineering needs.

[0006] The technical solution adopted by the present invention to solve the technical problem is as follows: In a first aspect, the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry, comprising: Get a dataset containing CSA cement retarders and corresponding molecular structures; Establishing a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor and a quantum chemical descriptor; Establish a quantitative structure-activity relationship model corresponding to each molecular descriptor and coagulation time; Based on the quantitative structure-activity relationship model, the molecular weight performance parameters of different retarders on the surface of CSA cement are calculated using quantum chemical calculation methods; According to the calculated molecular weight performance parameters of different retarders on the surface of CSA cement, the CSA cement retarders that meet the conditions are screened and output.

[0007] In one implementation, the step of establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time includes: Using the Lasso regression algorithm to respectively identify the features in the custom molecular composition descriptor, the Morgan fingerprint descriptor and the quantum chemical descriptor, and obtaining the Lasso CV model corresponding to each molecular descriptor according to the identified features; Based on mean square error and R 2 The parameters are used to comprehensively evaluate the Lasso CV model corresponding to each molecular descriptor, and the predetermined hyperparameters of the Lasso CV model corresponding to each molecular descriptor are selected according to the comprehensive evaluation results to obtain the quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time.

[0008] In one implementation, the mean square error and R 2 The parameters comprehensively evaluate the Lasso CV model corresponding to each molecular descriptor, including: The goodness of fit of the Lasso CV model corresponding to each descriptor is measured based on the minimum value of the mean square error, and the R 2 The parameter measures the degree of explanation of the data variability of the Lasso CV model corresponding to each descriptor, and obtains the comprehensive evaluation result.

[0009] In one implementation, selecting predetermined hyperparameters of the Lasso CV model corresponding to each molecular descriptor according to the comprehensive evaluation result includes: According to the comprehensive evaluation result, the model parameters corresponding to the custom molecular composition descriptor and the quantum chemical descriptor with adsorption energy and the hyperparameter values ​​of the selected feature quantity are selected as the predetermined hyperparameters of the corresponding Lasso CV model.

[0010] In one implementation, the molecular weight performance parameters of different retarders on the surface of CSA cement are calculated based on the quantitative structure-activity relationship model using a quantum chemical calculation method, including: Based on the quantitative structure-activity relationship model, the relationship between the weights corresponding to the carboxyl group, the ratio of P atoms to area, the ratio of hydroxyl groups to area, the custom functional groups, and the ratio of custom functional groups to area of ​​each retarder and the hyperparameters is analyzed to determine the characteristic values ​​and weights corresponding to each retarder; According to the corresponding characteristic values ​​and weights of each retarder, the molecular weight performance parameters of each retarder on the surface of CSA cement were calculated using the quantum chemical calculation method.

[0011] In one implementation, the molecular weight performance parameters of each retarder on the surface of CSA cement are calculated using a quantum chemical calculation method according to the characteristic values ​​and weights corresponding to each retarder, including: The van der Waals surface areas corresponding to the carboxyl segments of the molecular composition descriptors and the quantum chemical descriptors are calculated respectively to obtain the corresponding isosurface fractions of the electron density contributed by the hydroxy acid, so as to determine the molecular weight quantitative performance parameters of the corresponding retarder on the CSA cement surface.

[0012] In one implementation, the CSA cement retarder includes: one or a combination of citric acid, sodium citrate, tartaric acid, sodium tartrate, salicylic acid, malic acid, succinic acid, sodium gluconate, glycolic acid, sodium pyrophosphate, trisodium phosphate, disodium phosphate and sodium dihydrogen phosphate.

[0013] In a second aspect, the present invention provides a CSA cement retarder screening system based on machine learning and quantum chemistry, comprising: A data set acquisition module is used to acquire a data set including CSA cement retarder and corresponding molecular structure; A molecular descriptor establishment module, used to establish a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor and a quantum chemical descriptor; A quantitative structure-activity relationship model building module is used to build a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time; A quantum chemical calculation module, for calculating the molecular weight performance parameters of different retarders on the surface of CSA cement using a quantum chemical calculation method based on the quantitative structure-activity relationship model; The cement retarder screening module is used to screen and output the CSA cement retarders that meet the conditions according to the calculated molecular weight performance parameters of different retarders on the CSA cement surface.

[0014] In a third aspect, the present invention provides a terminal, comprising: a processor and a memory, wherein the memory stores a CSA cement retarder screening program based on machine learning and quantum chemistry, and when the CSA cement retarder screening program based on machine learning and quantum chemistry is executed by the processor, it is used to implement the operation of the CSA cement retarder screening method based on machine learning and quantum chemistry as described in the first aspect.

[0015] In a fourth aspect, the present invention further provides a medium, which is a computer-readable storage medium, and which stores a CSA cement retarder screening program based on machine learning and quantum chemistry. When the CSA cement retarder screening program based on machine learning and quantum chemistry is executed by a processor, it is used to implement the operation of the CSA cement retarder screening method based on machine learning and quantum chemistry as described in the first aspect.

[0016] The present invention adopts the above technical solution to achieve the following effects: (1) The present invention establishes a model by using quantum chemical descriptors, which can deeply analyze the interaction between retarder molecules and cement components and understand the action mechanism of retarder from a molecular level. In addition, with the help of quantum chemical descriptors, the machine learning model can quantify the influence of different molecules on the cement hydration process and predict the performance of the retarder under specific conditions, making the screening process more efficient and accurate. (2) The present invention uses a machine learning model to screen CSA cement retarders, which can learn a large amount of experimental data, predict the contribution of different components of the retarder to the retarding time, provide more accurate performance estimates, and help select appropriate retarder combinations in practical applications, thereby improving the quality stability of cement materials; (3) The present invention significantly accelerates the screening process of retarders by adopting a screening method that combines machine learning and quantum chemistry. It not only improves the screening efficiency and accuracy, but also reveals the action mechanism of the important components of the screened retarders through microscopic mechanisms, thereby improving the design and application capabilities of cement retarders. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of the CSA cement retarder screening method based on machine learning and quantum chemistry in the present invention.

[0019] Figure 2 It is a schematic diagram of the quantitative structure-activity relationship model of the three molecular descriptors in the present invention.

[0020] Figure 3 Schematic diagram of the Lasso CV model of three molecular descriptors in the present invention.

[0021] Figure 4It is a characteristic value screening diagram of the custom molecular composition descriptor and quantum chemical descriptor in the present invention.

[0022] Figure 5 It is a graph of the distribution and specific weights of the eigenvalues ​​selected in the present invention.

[0023] Figure 6 It is a schematic diagram of the calculation result of the surface area of ​​the retarder molecule ESP in the present invention.

[0024] Figure 7 Schematic diagram of the molecular electrostatic potential (MESP) value of the van der Waals surface in the present invention.

[0025] Figure 8 It is a functional principle diagram of a terminal in one implementation of the present invention.

[0026] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] Exemplary Methods Traditional retarders are screened mainly through experimental means to test the setting time, fluidity, early strength and other indicators of different additives in cement. The limitations of these screening methods mainly include technical defects such as long experimental cycles, limited condition control and lack of detailed understanding of reaction mechanisms. Due to the diversity of molecular structures and functions of retarders, traditional methods that rely on trial and error consume a lot of manpower and time costs in finding suitable reactant types and ratios, and it is difficult to accurately control the performance of retarders, and it is impossible to efficiently customize them to meet specific engineering needs.

[0029] In view of the above technical problems, an embodiment of the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry. The method can establish multiple molecular descriptors of CSA cement retarders according to the data set by acquiring a data set containing CSA cement retarders and corresponding molecular structures; wherein the multiple molecular descriptors include: custom molecular composition descriptors, Morgan fingerprint descriptors and quantum chemical descriptors; and by establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and the setting time, the molecular weight quantified performance parameters of different retarders on the surface of CSA cement can be calculated based on the quantitative structure-activity relationship model by using a quantum chemical calculation method; thereby, according to the calculated molecular weight quantified performance parameters of different retarders on the surface of CSA cement, the CSA cement retarders that meet the conditions are screened and output. Therefore, the embodiment of the present invention greatly accelerates the screening process of retarders by adopting a screening method combining machine learning and quantum chemistry, which not only improves the screening efficiency and accuracy, but also reveals the action mechanism of the important components of the screened retarders through microscopic mechanisms, thereby improving the design and application capabilities of cement retarders.

[0030] like Figure 1 As shown, an embodiment of the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry, comprising the following steps: Step S100, obtaining a data set including CSA cement retarder and corresponding molecular structure; Step S200, establishing a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor, and a quantum chemical descriptor.

[0031] In this example, the molecular level understanding is established by analyzing the influence of the molecular structure, electronic properties, reactivity and other factors of the retarder on the retarding effect of CSA cement; and the microscopic interaction between the retarder and the cement material is deeply explored through quantum chemical descriptors, thereby revealing the mechanism of action of the retarder.

[0032] In order to analyze the influencing factors of the retarding effect of CSA cement from the perspective of microscopic interactions, in this embodiment, it is necessary to first obtain a data set containing a variety of CSA cement retarders and corresponding molecular structures; wherein, the data set can be obtained from corresponding public chemical databases, such as PubChem, ChemSpider, etc., and these chemical databases are used to search for CSA cement retarders and their corresponding molecular structure information.

[0033] After obtaining a data set including CSA cement retarders and corresponding molecular structures, in this embodiment, three molecular descriptors of 13 organic small molecule retarders can be established according to the data set; wherein, the three molecular descriptors include: a custom molecular composition descriptor (MCD), a Morgan fingerprint descriptor (ECFP) and a quantum chemical descriptor (QCD); the custom molecular composition descriptor is used to quantify the structural characteristics or chemical properties of the custom molecular composition (i.e., the custom functional group molecular composition, such as hydroxyl, carboxyl, amino, aldehyde, etc.) of the CSA cement retarder; the Morgan fingerprint descriptor is used to describe the structure and similarity of the CSA cement retarder molecules; the quantum chemical descriptor is used to quantify the electronic properties of the CSA cement retarder molecules, which can be used to predict the physical and chemical properties of the molecules.

[0034] As an example, in this embodiment, 13 organic small molecule retarders including citric acid (CA), sodium citrate (SC), tartaric acid (TA), sodium tartrate (ST), salicylic acid (SA), malic acid (MA), succinic acid (BA), sodium gluconate (SG), glycolic acid (GC), sodium pyrophosphate (SP), trisodium phosphate (TP), disodium phosphate (DSP) and sodium dihydrogen phosphate (SDP) can be used as CSA cement retarders.

[0035] This embodiment establishes custom molecular composition descriptors, Morgan fingerprint descriptors, and quantum chemical descriptors. Custom molecular composition descriptors can be used to analyze the structural characteristics or chemical properties of custom functional groups in retarders, and Morgan fingerprint descriptors can be used to analyze the structure and similarity of molecules in retarders. Quantum chemical descriptors (such as molecular orbitals, polarizability, charge distribution, etc.) are used to deeply analyze the interaction between retarder molecules and cement components, and understand the mechanism of action of retarders from a molecular level. This method not only focuses on macroscopic effects, but also reveals the microscopic mechanism of action of retarders. In addition, with the help of quantum chemical descriptors, machine learning models can be used to quantify the effects of different molecules on the cement hydration process and predict the performance of retarders under specific conditions, making the screening process more efficient and accurate.

[0036] like Figure 1 As shown, an embodiment of the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry, comprising the following steps: Step S300, establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time.

[0037] In this embodiment, after establishing the custom molecular composition descriptor, Morgan fingerprint descriptor and quantum chemical descriptor, an effective machine learning model for sparse data can be constructed based on these three molecular descriptors, that is, an effective quantitative structure-activity relationship model (QSAR model) between various molecular descriptors and setting time is established. The QSAR model is an important tool in chemometrics, which uses mathematical statistics to explain the quantitative change law between the activity or physicochemical characteristics of a compound and its molecular structure. These quantitative structure-activity relationship models are used to analyze the relationship between the structure, functional group, quantum chemical properties and the retarding time of the CSA retarder, determine the molecular weight quantitative performance parameters of different retarders on the CSA cement surface, and thus screen and output the CSA cement retarder that meets the conditions.

[0038] Specifically, in an implementation of this embodiment, step S300 includes the following steps: Step S301, using the Lasso regression algorithm to respectively identify features in the custom molecular composition descriptor, the Morgan fingerprint descriptor, and the quantum chemical descriptor, and obtaining a Lasso CV model corresponding to each molecular descriptor according to the identified features; In this example, the main purpose is to build a quantitative structure-activity relationship model between the molecular descriptors of 13 retarders and the setting time; by using a separate machine learning model for each molecular descriptor, the descriptor most relevant to the setting time is determined. These descriptors include Morgan fingerprint descriptors, custom molecular composition descriptors, and quantitative descriptors (i.e., quantum chemical descriptors) based on density functional theory (DFT) calculations, such as Figure 2 shown.

[0039] While ensuring the accuracy of the three types of molecular descriptor models, this embodiment uses the LASSO method (i.e., the Lasso regression algorithm, a method used for regression analysis in statistics and machine learning, which constrains the complexity of the model by adding a regularization term) to identify the most relevant features in each descriptor type. The purpose is to verify the importance of multidimensional descriptors through this descriptor combination method. Since the descriptor data in this embodiment has multiple characteristic attributes, a standard normalization procedure is applied in the calculation process in this embodiment. Integrating normalization can enhance the learning ability of the model and improve its accuracy to a certain extent.

[0040] In this embodiment, the corresponding Lasso CV model diagram is obtained based on the Morgan fingerprint descriptor, the custom molecular composition descriptor and the quantum chemical descriptor. Figure 3 As shown, Figure 3 (a) is the Lasso CV model diagram corresponding to the Morgan fingerprint descriptor; Figure 3 (b) is the Lasso CV model diagram corresponding to the custom molecular composition descriptor; Figure 3(c) is the Lasso CV model diagram corresponding to the quantum chemical descriptor.

[0041] Step S302: based on the mean square error and R 2 The parameters are used to comprehensively evaluate the Lasso CV model corresponding to each molecular descriptor, and the predetermined hyperparameters of the Lasso CV model corresponding to each molecular descriptor are selected according to the comprehensive evaluation results to obtain the quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time.

[0042] In this embodiment, for the Lasso CV model of each molecular descriptor, the performance of the model can be comprehensively evaluated by two indicators, MSE (mean square error) and R2 (goodness of fit). Among them, the minimum value of MSE is used to measure the goodness of fit of the model, and R2 is used to measure the degree to which the model explains the variability of the data. Then, the best hyperparameters (i.e., predetermined hyperparameters) are selected by cross-validation. Here, function to obtain the best model performance.

[0043] Specifically, in an implementation of this embodiment, step S302 includes the following steps: Step S302a, measuring the goodness of fit of the Lasso CV model corresponding to each descriptor based on the minimum value of the mean square error, and 2 The parameter measures the degree of explanation of the data variability of the Lasso CV model corresponding to each descriptor, and obtains the comprehensive evaluation result.

[0044] In this embodiment, the Lasso CV model of the Morgan fingerprint descriptor is as follows: Figure 3 As shown in (a). When R2 is -9.81, it reaches its maximum value and the corresponding MSE is its minimum value. However, 0.157 is much smaller than 0.50, so the LassoCV model has no model effect on descriptors such as the Morgan fingerprint descriptor.

[0045] The Lasso CV models of custom molecular composition descriptors are shown below: Figure 3 As shown in (b). As the MSE increases, it continues to gradually decrease until it reaches a plateau of 0.354. At the same time, R2 shows an upward trend, slows down at this point, and finally stabilizes. In order to enhance the Lasso CV model to capture the relationship between the custom molecular composition descriptor and the delay time, it is crucial to achieve a sufficiently high R2 while minimizing the MSE. Finally, it is determined that The value is set to -2.32, corresponding to When the value is 0.098, the model works best.

[0046] The Lasso CV model corresponding to the quantum chemical descriptor is shown in the figure Figure 3 As shown in (c), Figure 3 (c) compares the model effects before and after adding adsorption energy; among them, curves 1 and 3 are the effects without adding adsorption energy, and curves 2 and 4 are the effects with adding adsorption energy. It can be seen that the quantum chemical descriptor is obvious for the stability of the model. When R 2 After reaching the maximum value, MSE also reaches the minimum value, and the model is finally stable. Obviously, after adding adsorption energy, the stability of the model R 2 It increased from 0.335 to 0.942, and this huge increase is a great endorsement of the adsorption energy descriptor. In addition, the Lasso CV model achieved the best screening effect.

[0047] In this embodiment, based on the above goodness of fit and the degree of explained data variability, after comprehensive analysis, the Lasso CV model corresponding to the custom molecular composition descriptor and the quantum chemical descriptor is selected as the model for explaining molecular quantum chemical information.

[0048] Specifically, in an implementation of this embodiment, step S302 further includes the following steps: Step S302b, according to the comprehensive evaluation result, select the model parameters corresponding to the custom molecular composition descriptor and the quantum chemical descriptor with adsorption energy and the hyperparameter values ​​of the selected feature quantity as the predetermined hyperparameters of the corresponding Lasso CV model.

[0049] In this example, to evaluate the LASSO model of molecular descriptors, MSE, RMSE (root mean square error), MAE (squared absolute error), and R 2 By comparing QMD-1 (quantum chemical descriptor model 1) without adsorption energy with QMD-2 (quantum chemical descriptor model 2) with adsorption energy, QMD-2 performs best in MSE and QMD-1 performs worst in MSE. RMSE and MAE also show the same trend in these two models. Obviously, QMD-2 is the best choice model to explain the molecular quantum chemical information by increasing the adsorption energy. Values, number of selected features, MSE, RMSE, MAE, and R 2 As shown in Table 1.

[0050] Table 1. Hyperparameter values ​​of descriptors, model parameters, and number of selected features.

[0051] The model results show that when the optimal The Morgan fingerprint descriptor screens 15 feature descriptors when the value is , which is not very meaningful for screening important components of a small data set. Both the quantum chemical descriptor and the custom molecular composition descriptor show promising ‌QSAR results, and the R 2 The values ​​are 0.942 and 0.602, respectively, and the MAE values ​​are 0.215 and 0.551. This shows that both models are important for screening the descriptors of the delay time. When the value is set to minimize the uncertainty in the MSE of QMD-2, the model becomes sparse, and the five key parameters that have the most significant impact on the condensation time of the molecular pair are retained. Therefore, in this embodiment, based on the results of the comprehensive evaluation, the model parameters corresponding to the custom molecular composition descriptor and the quantum chemical descriptor with adsorption energy and the hyperparameter values ​​of the number of selected features are selected as the optimal hyperparameters of the corresponding Lasso CV model.

[0052] In this embodiment, the five screened quantum chemical descriptors are combined with the five selected custom molecular composition descriptors through the above cross-validation method, and the molecular properties can be integrated to better understand the mechanism by which the molecular composition of the blocker affects the delay time.

[0053] This embodiment constructs a quantitative structure-activity relationship model corresponding to each molecular descriptor and the setting time, and learns a large amount of experimental data in the form of a machine learning model, which can predict the contribution of different components of the retarder to the retarding time. Compared with traditional methods, the prediction model based on machine learning can provide more accurate performance estimates, help users select suitable retarder combinations in practical applications, thereby improving the quality stability of cement materials and guiding the synthesis of new materials. And through data-driven methods, it can reduce experimental costs and resource consumption, optimize the R&D process, thereby reducing the number of experiments and optimizing the selection of the best retarder components.

[0054] like Figure 1 As shown, an embodiment of the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry, comprising the following steps: Step S400, based on the quantitative structure-activity relationship model, a quantum chemical calculation method is used to calculate the molecular weight performance parameters of different retarders on the surface of CSA cement.

[0055] In this example, based on the established quantitative structure-activity relationship model, the cement system is comprehensively considered, and the quantum chemical calculation method is used to calculate the adsorption properties of different retarders on the CSA surface from an atomic perspective. The contribution and mechanism of the key components of the retarder to cement hydration are studied.

[0056] Specifically, in an implementation of this embodiment, step S400 includes the following steps: Step S401, based on the quantitative structure-activity relationship model, the relationship between the weights corresponding to the carboxyl group, the ratio of P atoms to area, the ratio of hydroxyl groups to area, the custom functional groups, and the ratio of custom functional groups to area of ​​each retarder and the hyperparameters is analyzed to determine the eigenvalues ​​and weights corresponding to each retarder.

[0057] In this embodiment, when analyzing the relationship between the structure, functional groups, quantum chemical properties and setting time of CSA cement retarder based on the quantitative structure-activity relationship model, it is also necessary to determine the weight of CSA cement retarder; the purpose of determining the weight is to filter out the number of eigenvalues ​​while ensuring the interpretability of the model and determining the hyperparameters , in order to filter out more unimportant parameters.

[0058] like Figure 4 As shown, in this embodiment, R with characteristic values ​​is plotted 2 Trend curve, the shaded area is the number of eigenvalues ​​in the final calculated (0-1) range. The dashed line represents the point where the model is most reliable, at which the number of eigenvalues ​​should be less than the number of samples. Figure 4 (a) shows the process of determining the feature value of the custom molecular composition descriptor. 2 0.602, and 5 eigenvalues ​​were initially selected in this embodiment, which is obviously inconsistent with the intended use of the Lasso CV model. In the shaded range, the feature range selected in this embodiment is less than 13, and the MSE meets the minimum value. Therefore, the 5 eigenvalues ​​screened by this model are sufficient to filter out unimportant features.

[0059] Figure 4 (b) is the eigenvalue screening diagram of the QMD-2 quantitative descriptor model with increased adsorption energy. 2 It shows a general trend of increasing first and then decreasing. Figure 4 As shown in (b), this embodiment only considers the eigenvalues ​​selected when the model is reliable. Select R 2 The inflection point, i.e. the value at the highest point of the curve. The quantitative properties of the five molecules were determined by the model, which explained the effect on the slow setting time.

[0060] Previous studies have usually focused on the interpretability of models and the selection of parameters to account for the predictability of applied models. Figure 5 As shown, in this embodiment, Figure 5 A task to study the weights of selected features was added to investigate the importance of custom molecular composition descriptors and quantum chemical descriptor properties in influencing the slow setting time. It is believed that molecular composition plays a crucial role in determining the molecular properties. This work also helps in investigating the mechanism of molecular composition in slow setting time in this example.

[0061] like Figure 5 As shown in (a), this example explores the relationship between the weights of COOH (carboxyl), P / MV (ratio of P atoms to area), OH / MV (ratio of hydroxyl groups to area), R (custom functional group) and R / MV (ratio of custom functional group to area) and hyperparameters. In fact, these five descriptors were selected from the best case of the previous model. As the hyperparameters increase, the weights of the descriptors continue to decrease, and finally the weight of each descriptor is -2.32. OK. R (custom functional group) and COOH have the highest weights, 0.596 and 0.482, respectively. Note that P also occupies an important proportion, which indicates that P atoms also make important contributions to the molecular structure and properties.

[0062] like Figure 5 As shown in (b), the best contributing quantum chemical descriptor selected by the Lasso CV model is shown. It is found that the molecular composition, especially COOH and P, affects the role of the molecule in the reaction. QMD-2 found that the van der Waals specific surface area has a significant effect on the reaction properties of the molecule. It is worth noting that the important indicator of adsorption energy must also be considered as an important quantitative descriptor in the reaction system.

[0063] Specifically, in an implementation of this embodiment, step S400 further includes the following steps: Step S402, according to the characteristic values ​​and weights corresponding to each retarder, the molecular weight performance parameters of each retarder on the surface of CSA cement are calculated using a quantum chemical calculation method.

[0064] The method of using a quantum chemical calculation method to calculate the molecular weight quantified performance parameters of each retarder on the surface of CSA cement according to the characteristic values ​​and weights corresponding to each retarder comprises: respectively calculating the van der Waals surface area of ​​the molecular composition descriptor and the quantum chemical descriptor corresponding to the carboxyl fragment, and obtaining the corresponding isosurface fraction of the electron density contributed by the hydroxy acid, so as to determine the molecular weight quantified performance parameters of the corresponding retarder on the surface of CSA cement.

[0065] In this example, custom molecular composition descriptors and quantum chemical descriptors provide important eigenvalues ​​for the effect on coagulation time, especially COOH, while A is an important descriptor for differences in molecular properties. Calculating the van der Waals surface area corresponding to the COOH fragment can quantify the fraction of the isosurface with an electron density of 0.001 au contributed by COOH[R] (hydroxy acid).

[0066] like Figure 6 As shown, in Figure 6In the figure, each point corresponds to a peak on the electron density isosurface, and the blue area corresponds to the COOH fragments in the nine acidic retarders. The blue and red parts represent the COOH functional group and other functional groups, respectively. The van der Waals area occupied by COOH is shown in Figure 6 Comparison of the GC (glycolic acid) molecule shows that the carboxyl functional group is equivalent to about twice the van der Waals surface area of ​​the hydroxyl group, accounting for 46.53% of the total area. The COOH groups on other molecules also account for more than 23.92% of the total fraction. This suggests that a high percentage of COOH (reflecting the strength of the interaction between the molecule and the solid surface) plays an important role in promoting the adsorption behavior.

[0067] like Figure 1 As shown, an embodiment of the present invention provides a CSA cement retarder screening method based on machine learning and quantum chemistry, comprising the following steps: Step S500, based on the calculated molecular weight performance parameters of different retarders on the surface of CSA cement, screen and output the CSA cement retarders that meet the conditions.

[0068] In this embodiment, according to the calculated molecular weight performance parameters of different retarders on the surface of CSA cement, CSA cement retarders meeting the ESP value range are screened and output.

[0069] In this example, the quantitative molecular surface analysis focuses on the distribution of the electrostatic potential (ESP) of the phosphate blocker molecule on the molecular van der Waals surface. The Bader method (an intuitive method for decomposing molecules into atoms) is still used to define the molecular van der Waals surface, and an equivalent surface with an electron density of 0.001 e / bohr^3 is used as the molecular van der Waals surface to reflect the effects of local electronic features. Figure 7 , surface area histograms for different ESP ranges, and electrostatic potential maps of the equivalent surfaces of the molecules are shown. In these maps, red indicates more negative electrostatic potential, while blue indicates more positive potential.

[0070] Figure 7 The ESP splitting of the entire phosphate retarder molecule is shown in Figure 1. Figure 7It can be clearly seen that the surface area distribution of the molecule in different electrostatic potential intervals is uneven, and the negative area occupies most of the molecular surface. The ESP values ​​of SP, TP, and DSP are -70kcal / mol - 35kcal / mol, and the ESP value of SDP is -50kcal / mol - 25kcal / mol, which is mainly due to the obvious negative charge of the carboxyl and phosphate groups on oxygen. It can also be observed that the very positive values ​​in the blue region are mainly due to the generation of Na (105kcal / mol - 140kcal / mol). The hydroxyl, carboxyl and Na groups on the phosphate retarder cause the molecule to have different electrostatic properties. The vdW surface electrostatic potential has been widely used to study the molecular reaction sites under test, the sites with stronger electronegativity, i.e. Figure 7 The red areas in the graph may become the molecular reaction sites. The results show that the functional groups on the molecule are the key indicators of molecular reactivity, and the van der Waals area calculated by the quantitative descriptor is mainly affected by the carboxyl group. The characteristics of these functional groups will significantly affect the reaction of the retarder in a specific environment, thereby affecting the setting time.

[0071] This embodiment uses existing experimental data to train a machine learning model in order to predict the performance of new compounds, thereby reducing the need for repeated experiments. Through the machine learning algorithm, candidate agents with excellent retarding effects are quickly identified and screened, significantly shortening the time and cost required for traditional experimental screening. In addition, the screening process of retarders is greatly accelerated by the screening method combining machine learning and quantum chemistry. The method for screening cement retarders based on machine learning and quantum chemical descriptors not only improves the screening efficiency and accuracy, but also reveals the mechanism of action of the important components of the screened retarders through microscopic mechanisms. This technology gives the advantage of data-driven research and development of retarders, promotes the intelligent research and development of cement materials, and forms an efficient and accurate material design and application system. Integrate machine learning and materials science, establish an intelligent material research and development platform, and improve the design and application capabilities of cement retarders.

[0072] This embodiment achieves the following technical effects through the above technical solution: (1) This embodiment uses quantum chemical descriptors to establish a model, which can deeply analyze the interaction between retarder molecules and cement components and understand the mechanism of action of the retarder from a molecular level. In addition, with the help of quantum chemical descriptors, the machine learning model can quantify the effects of different molecules on the cement hydration process and predict the performance of the retarder under specific conditions, making the screening process more efficient and accurate. (2) This embodiment uses a machine learning model to screen CSA cement retarders, which can learn a large amount of experimental data, predict the contribution of different components of the retarder to the retarding time, provide more accurate performance estimation, and help select a suitable retarder combination in practical applications, thereby improving the quality stability of cement materials; (3) This embodiment greatly accelerates the screening process of retarders by adopting a screening method that combines machine learning and quantum chemistry. It not only improves the screening efficiency and accuracy, but also reveals the action mechanism of the important components of the screened retarders through microscopic mechanisms, thereby improving the design and application capabilities of cement retarders.

[0073] Exemplary Devices Based on the above embodiments, the present invention further provides a CSA cement retarder screening system based on machine learning and quantum chemistry, comprising: A data set acquisition module is used to acquire a data set including CSA cement retarder and corresponding molecular structure; A molecular descriptor establishment module, used to establish a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor and a quantum chemical descriptor; A quantitative structure-activity relationship model building module is used to build a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time; A quantum chemical calculation module, for calculating the molecular weight performance parameters of different retarders on the surface of CSA cement using a quantum chemical calculation method based on the quantitative structure-activity relationship model; The cement retarder screening module is used to screen and output the CSA cement retarders that meet the conditions according to the calculated molecular weight performance parameters of different retarders on the CSA cement surface.

[0074] This embodiment achieves the following technical effects through the above technical solution: (1) This embodiment uses quantum chemical descriptors to establish a model, which can deeply analyze the interaction between retarder molecules and cement components and understand the mechanism of action of the retarder from a molecular level. In addition, with the help of quantum chemical descriptors, the machine learning model can quantify the effects of different molecules on the cement hydration process and predict the performance of the retarder under specific conditions, making the screening process more efficient and accurate. (2) This embodiment uses a machine learning model to screen CSA cement retarders, which can learn a large amount of experimental data, predict the contribution of different components of the retarder to the retarding time, provide more accurate performance estimation, and help select a suitable retarder combination in practical applications, thereby improving the quality stability of cement materials; (3) This embodiment greatly accelerates the screening process of retarders by adopting a screening method that combines machine learning and quantum chemistry. It not only improves the screening efficiency and accuracy, but also reveals the action mechanism of the important components of the screened retarders through microscopic mechanisms, thereby improving the design and application capabilities of cement retarders.

[0075] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be as follows: Figure 8 shown.

[0076] The terminal includes: a processor, a memory, an interface, a display screen and a communication module connected through a system bus; wherein the processor of the terminal is used to provide computing and control capabilities; the memory of the terminal includes a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and the computer program in the storage medium; the interface is used to connect to external devices; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or other devices.

[0077] The computer program is used to implement the operation of the CSA cement retarder screening method based on machine learning and quantum chemistry when executed by a processor.

[0078] It can be understood by those skilled in the art that Figure 8 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0079] In one embodiment, a terminal is provided, which includes: a processor and a memory, wherein the memory stores a CSA cement retarder screening program based on machine learning and quantum chemistry, and the CSA cement retarder screening program based on machine learning and quantum chemistry is used to implement the operations of the above-mentioned CSA cement retarder screening method based on machine learning and quantum chemistry when the processor executes the CSA cement retarder screening program based on machine learning and quantum chemistry.

[0080] In one embodiment, a storage medium is provided, wherein the storage medium stores a CSA cement retarder screening program based on machine learning and quantum chemistry, and when the CSA cement retarder screening program based on machine learning and quantum chemistry is executed by a processor, it is used to implement the operation of the above-mentioned CSA cement retarder screening method based on machine learning and quantum chemistry.

[0081] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and volatile memory.

[0082] In summary, the present invention provides a method for screening CSA cement retarders based on machine learning and quantum chemistry, including: obtaining a data set containing CSA cement retarders and corresponding molecular structures; establishing multiple molecular descriptors of CSA cement retarders according to the data set; wherein the multiple molecular descriptors include: custom molecular composition descriptors, Morgan fingerprint descriptors, and quantum chemical descriptors; establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and setting time; based on the quantitative structure-activity relationship model, using a quantum chemical calculation method to calculate the molecular weight quantified performance parameters of different retarders on the surface of CSA cement; based on the calculated molecular weight quantified performance parameters of different retarders on the surface of CSA cement, screening and outputting CSA cement retarders that meet the conditions. The present invention improves the accuracy and efficiency of cement retarder screening.

[0083] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A CSA cement retarder screening method based on machine learning and quantum chemistry, characterized in that: include: Get a dataset containing CSA cement retarders and corresponding molecular structures; Establishing a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor and a quantum chemical descriptor; Establish a quantitative structure-activity relationship model corresponding to each molecular descriptor and coagulation time; Based on the quantitative structure-activity relationship model, the molecular weight performance parameters of different retarders on the surface of CSA cement are calculated using quantum chemical calculation methods; According to the calculated molecular weight performance parameters of different retarders on the surface of CSA cement, the CSA cement retarders that meet the conditions are screened and output.

2. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 1, characterized in that: The method of establishing a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time includes: Using the Lasso regression algorithm to respectively identify the features in the custom molecular composition descriptor, the Morgan fingerprint descriptor and the quantum chemical descriptor, and obtaining the Lasso CV model corresponding to each molecular descriptor according to the identified features; Based on mean square error and R 2 The parameters are used to comprehensively evaluate the Lasso CV model corresponding to each molecular descriptor, and the predetermined hyperparameters of the Lasso CV model corresponding to each molecular descriptor are selected according to the comprehensive evaluation results to obtain the quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time.

3. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 2, characterized in that: The mean square error and R 2 The parameters comprehensively evaluate the Lasso CV model corresponding to each molecular descriptor, including: The goodness of fit of the Lasso CV model corresponding to each descriptor is measured based on the minimum value of the mean square error, and the R 2 The parameter measures the degree of explanation of the data variability of the Lasso CV model corresponding to each descriptor, and obtains the comprehensive evaluation result.

4. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 2, characterized in that: The predetermined hyperparameters of the Lasso CV model corresponding to each molecular descriptor are selected according to the comprehensive evaluation results, including: According to the comprehensive evaluation result, the model parameters corresponding to the custom molecular composition descriptor and the quantum chemical descriptor with adsorption energy and the hyperparameter values ​​of the selected feature quantity are selected as the predetermined hyperparameters of the corresponding Lasso CV model.

5. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 1, characterized in that: Based on the quantitative structure-activity relationship model, the molecular weight performance parameters of different retarders on the surface of CSA cement are calculated using a quantum chemical calculation method, including: Based on the quantitative structure-activity relationship model, the relationship between the weights corresponding to the carboxyl group, the ratio of P atoms to area, the ratio of hydroxyl groups to area, the custom functional groups, and the ratio of custom functional groups to area of ​​each retarder and the hyperparameters is analyzed to determine the characteristic values ​​and weights corresponding to each retarder; According to the corresponding characteristic values ​​and weights of each retarder, the molecular weight performance parameters of each retarder on the surface of CSA cement were calculated using the quantum chemical calculation method.

6. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 5, characterized in that: According to the characteristic values ​​and weights corresponding to each retarder, the molecular weight performance parameters of each retarder on the surface of CSA cement are calculated using a quantum chemical calculation method, including: The van der Waals surface areas corresponding to the carboxyl segments of the molecular composition descriptors and the quantum chemical descriptors are calculated respectively to obtain the corresponding isosurface fractions of the electron density contributed by the hydroxy acid, so as to determine the molecular weight quantitative performance parameters of the corresponding retarder on the CSA cement surface.

7. The CSA cement retarder screening method based on machine learning and quantum chemistry according to claim 1, characterized in that: The CSA cement retarder comprises: one or a combination of citric acid, sodium citrate, tartaric acid, sodium tartrate, salicylic acid, malic acid, succinic acid, sodium gluconate, glycolic acid, sodium pyrophosphate, trisodium phosphate, disodium phosphate and sodium dihydrogen phosphate.

8. A CSA cement retarder screening system based on machine learning and quantum chemistry, characterized in that: include: A data set acquisition module is used to acquire a data set including CSA cement retarder and corresponding molecular structure; A molecular descriptor establishment module, used to establish a plurality of molecular descriptors of the CSA cement retarder according to the data set; wherein the plurality of molecular descriptors include: a custom molecular composition descriptor, a Morgan fingerprint descriptor and a quantum chemical descriptor; A quantitative structure-activity relationship model building module is used to build a quantitative structure-activity relationship model corresponding to each molecular descriptor and the coagulation time; A quantum chemical calculation module, for calculating the molecular weight performance parameters of different retarders on the surface of CSA cement using a quantum chemical calculation method based on the quantitative structure-activity relationship model; The cement retarder screening module is used to screen and output the CSA cement retarders that meet the conditions according to the calculated molecular weight performance parameters of different retarders on the CSA cement surface.

9. A terminal, characterized in that: include: A processor and a memory, wherein the memory stores a CSA cement retarder screening program based on machine learning and quantum chemistry, and when the CSA cement retarder screening program based on machine learning and quantum chemistry is executed by the processor, it is used to implement the operation of the CSA cement retarder screening method based on machine learning and quantum chemistry as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a CSA cement retarder screening program based on machine learning and quantum chemistry, and when the CSA cement retarder screening program based on machine learning and quantum chemistry is executed by a processor, it is used to implement the operation of the CSA cement retarder screening method based on machine learning and quantum chemistry as described in any one of claims 1 to 7.

Images