A Machine Learning-Based Molecular Design Method for Ethylene Oligomerization Catalysts

Abstract

This invention discloses a machine learning-based molecular design method for ethylene oligomerization catalysts, comprising: extracting descriptors of the catalytic system as independent variables and experimental or calculated oligomerization results as dependent variables to establish a database of the catalytic system; establishing a model relationship between the descriptors and experimental or calculated oligomerization results through database learning to obtain a machine learning model; and predicting the target catalytic system and its ligands based on the target oligomerization results according to the machine learning model. The machine learning-based ethylene oligomerization design method of this invention, applied to ethylene trimerization and tetramerization, shows a high degree of agreement between experimental and target values, providing a new approach for rapidly designing the required catalytic system and its ligands, thereby enabling rapid screening and highly accurate prediction of highly selective catalytic systems.

Description

Technical Field

[0001] This invention belongs to the field of catalyst molecular design, and in particular relates to a method for designing molecular catalysts for ethylene oligomerization based on machine learning. Background Technology

[0002] Linear alpha-olefins are important polyolefin comonomers and crucial intermediates in the production of high-value chemicals such as plasticizers, detergents, surfactants, cosmetics, and lubricants. Polyolefin products produced using 1-hexene and 1-octene as comonomers exhibit good mechanical properties and excellent processing performance, making them widely in demand in the polyolefin industrial production.

[0003] Currently, titanium-based, tantalum-based, and chromium-based catalysts have all exhibited good catalytic performance in the selective oligomerization of ethylene. Among them, Cr-based catalysts have superior comprehensive performance (activity and selectivity) and have therefore attracted much attention. Research on Cr-based catalysts for ethylene oligomerization can be traced back to the 1970s. Manyik et al. first reported in 1977 that the Cr(III)2-EH / PIBAO system could catalyze the selective trimerization of ethylene, with 1-hexene as the main component of the product [J Catal. 47, 197 (1977)]. In 1989, Briggs reported the chromium-catalyzed selective trimerization of ethylene to produce 1-hexene [Chem. Commun. 674 (1989)]. Subsequently, many scholars have also shown great interest in the chromium-based catalysts for the trimerization of ethylene to 1-hexene [EP0417477; US5811618; Chem. Commun. 858 (2002); AdV. Synth. Catal. 345, 1 (2003); Appl. Catal. A 255, 355 (2003); Chem. Commun. 334 (2003); J. Am. Chem. Soc. 125, 5272 (2003); J. Organomet. Chem. 690, 713 (2005); Chem. Commun. 620 (2005); Organometallics 25, 3605 (2006)]. Research hotspots mainly focus on bidentate phosphine ligands [J.Am.Chem.Soc.126,14712(2004); Chem.Commun.622(2005); Appl.Catal.A 306,184(2006); AdV.Synth.Catal.348,1200(2006)], SNS [Chem.Commun.334(2003); Organometallics 24,552(2005)], and PPN [Eur.J.Inorg.Chem.2004,530(2004); Inorg.Chem.43,2228(2004); Coord.Chem.Rev.249,2056(2005)]. Most recent studies have also extended from this foundation [J.Am.Chem.Soc.141,6022(2019); ACS Catal.3,2582(2013); ACS Catal.10,9674(2020); Organometallics 36,1640(2017); ACS Catal.3,95(2013); J.Am.Chem.Soc.126,14712(2004); Organometallics 39,976(2020)].

[0004] One of the biggest challenges in the selective oligomerization of ethylene is the targeted control of product selectivity. While industry has achieved selectivity exceeding 99% for 1-hexene by adjusting co-catalysts and reaction conditions such as temperature and pressure, it still falls short of achieving high selectivity (>90%) for 1-octene. This control method is demanding, lacks controllability, and is difficult to implement at the molecular level. Industry and academia have invested significant resources in chromium-based catalytic systems, publishing numerous papers and patents, but progress in the high-selectivity preparation of 1-octene remains slow. This is mainly due to two factors: the complex conformations of reaction intermediates and key transition states, requiring precise structural analysis for the race-like steps of ethylene trimerization and tetramerization; and the complex mechanisms of ligand electronic and steric hindrance effects, necessitating precise theoretical guidance for catalyst molecular design. Summary of the Invention

[0005] To address the problems of inaccurate guidance for catalyst molecular design in the prior art, this invention provides a machine learning-based method for designing ethylene oligomerization catalysts, comprising:

[0006] Extract the descriptors of the catalytic system as independent variables, and the experimental or calculated oligomerization results as dependent variables to establish a database of the catalytic system;

[0007] By learning from the database, a model relationship is established between the descriptors and the convergence results of the experiments or calculations, thereby obtaining a machine learning model;

[0008] Based on the machine learning model, the target catalytic system and its ligands are predicted based on the target oligomerization results.

[0009] Preferably, the process of extracting the descriptor of the catalytic system as the independent variable and the experimental or calculated oligomerization result as the dependent variable, and establishing a database of the catalytic system includes:

[0010] A catalytic system was established using simulation software. The low-energy conformational structures of the key intermediates and transition states of the catalytic system were obtained based on conformational search. The relationship between the energy barrier of the rate-controlling step and the experimental results was compared and analyzed to extract the descriptor of the catalytic system. Using the descriptor of the catalytic system as the independent variable and the corresponding experimental results as the dependent variable, the relationship between the descriptor and the corresponding experimental results was obtained. Based on the relationship, a database of the catalytic system was constructed.

[0011] Preferably, after establishing the database of the catalytic system, the method further includes preprocessing the database;

[0012] The preprocessing process includes removing obvious errors or outliers from the data, and then normalizing and regularizing the data to serve as sample data for training the machine learning model.

[0013] Preferably, the catalytic system includes a Cr catalyst and a ligand;

[0014] The ligands include PNP ligands, PCCP ligands, PNSiP ligands, PNSiNP ligands, PNPN ligands, or (2-C5H4N)2NR ligands;

[0015] R is hydrogen or an organic group, including R1 group, R2 group, R3 group, R4 group, R5 group, R2' group or R3' group.

[0016] Preferably, the PNP ligand structure is as follows:

[0017]

[0018] The PCCP ligand structure is as follows:

[0019]

[0020] The structural formula of the PNSiP ligand is:

[0021]

[0022] The structural formula of the PNSiNP ligand is:

[0023]

[0024] The structural formula of the PNPN ligand is:

[0025]

[0026] The structural formula of the (2-C5H4N)2NR ligand is:

[0027]

[0028] Wherein, the ligand structural formula is:

[0029] The R1 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitrosyl, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl.

[0030] The R2 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl.

[0031] The R3 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl.

[0032] The R4 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl.

[0033] The R5 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl.

[0034] The R2' group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl.

[0035] The R3' group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl.

[0036] Preferably, the descriptor includes the bond length, bond angle, and dihedral angle of each skeletal atom of the Cr catalyst and ligand;

[0037] The oligomerization results are the coselectivity of 1-hexene and 1-octene and the reactivity of the Cr catalyst.

[0038] Preferably, the process of learning and establishing a model relationship between the descriptors and the convergence results of experiments or calculations through the database to obtain a machine learning model includes:

[0039] Based on the database settings and training regression variables, the dataset is randomly divided into training and validation sets in a ratio of 30 / 70 to 90 / 10. The training set is then trained using a machine learning algorithm, and the hyperparameter combination with the best predictive performance in the training set is obtained through cross-validation. An initial machine learning model is then constructed based on the hyperparameter combination.

[0040] The data in the validation set is input into the initial machine learning model for validation to obtain the model prediction performance results;

[0041] The optimal machine learning model is selected based on the predicted performance results of the model, and the machine learning model is obtained.

[0042] Preferably, the machine learning algorithm is selected from two or more regression algorithms such as ridge regression, support vector machine, nearest neighbor algorithm, Bayesian algorithm, decision tree algorithm, neural network algorithm, random forest algorithm, Gaussian process regression algorithm, lasso algorithm, and elastic network algorithm.

[0043] The present invention discloses the following technical effects:

[0044] This invention provides a machine learning-based molecular design method for ethylene oligomerization catalysts, applied to ethylene trimerization and tetramerization. Experimental values ​​show a high degree of agreement with target values. This method overcomes the shortcomings of traditional ethylene oligomerization reactions, such as complex transition state and intermediate conformations, and complex mechanisms of ligand electronic effects and steric hindrance. It replaces traditional trial-and-error catalyst design methods, providing a new approach for rapidly designing desired catalytic systems and their ligands, thereby enabling rapid screening and highly accurate prediction of highly selective catalytic systems. Experimental verification shows that the ligands screened in this invention can significantly improve the co-selectivity of 1-hexene and 1-octene, thereby enhancing reactivity. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0047] Figure 2 The figure shows the data fitting results of the Cr-PNP system in an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1 As shown, this invention provides a method for molecular design of ethylene oligomerization catalysts based on machine learning, comprising:

[0051] Part 1: Obtaining Sample Data from the Ethylene Oligomerization Database

[0052] Example 1

[0053] The Cr-PNP trimer and tetramer system of ethylene has the following structural formula (1):

[0054]

[0055] The R1 group is selected from hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitrosyl, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (diisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl The R2 group is selected from one or more of the following groups: 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, 1-cyclohexyl, etc.; the R2 group is selected from one or more of the following groups: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, etc.

[0056] When the R2 group is phenyl, the structural formulas of the PNP type ligands are shown in formulas (7)-(10):

[0057]

[0058] The X group is selected from one or more groups including hydrogen, fluorine, iodo, methyl, ethyl, propyl, trichloromethyl, trifluoromethyl, methoxy, acetamido, toluamide, benoxy, phenoxy, and sulfonic acid groups.

[0059] Example of PNP ligand synthesis: At 0°C, diphenylphosphine chloride and the corresponding (isopropylcyclohexyl)methylamine were added to a dichloromethane solution at a molar ratio of 2:1 and stirred for 40 min, then stirred overnight at room temperature. The amine hydrochloride produced during the reaction was removed by filtration. The product was obtained by recrystallization with a yield of 81%.

[0060] Example of a reaction in the Cr-PNP system: Ethylene oligomerization was carried out in a 300 mL autoclave. Before the oligomerization began, the autoclave was purged with nitrogen three times and ethylene twice. Toluene solvent was then added to the autoclave, stirred under an ethylene atmosphere, and heated to the set temperature. Pre-determined amounts of methylaluminoxane (MAO), PNP ligands, and CrCl3(THF)3 were added to the autoclave and stirred for 1 min. Ethylene was introduced into the autoclave, and the pressure was adjusted to the set value, initiating the reaction. After 30 min of reaction, the ethylene gas was shut off, and the reaction system was rapidly cooled to 20 °C. The molar ratio of MAO, ligands, and CrCl3(THF)3 was 300:3:1, the reaction conditions were 60 °C, 4.5 MPa, and the amount of Cr added per reaction was 2.5 μmol. The experimental products were analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results of the experiment included: a coselectivity of 0.918 for 1-hexene and 1-octene, and a catalyst activity of 824703 g / (gCr h).

[0061] Similarly, 10 PNP-type ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0062] Example 2

[0063] The structural formula of the PCCP type ligand in the ethylene trimer and tetramer system Cr-PCCP is shown in the following formula (2):

[0064]

[0065] The selection range of R1 and R2 groups is consistent with that in formula (1); R3 group is selected from one or more of the following groups: hydrogen group, methyl group, ethyl group, propyl group, butyl group, cyclohexyl group, phenyl group, allyl group, tolyl group, formyl group, acetyl group, benzoyl group, nitro group, nitrosyl group, fluorine group, bromo group, iodo group, chloro group, amino group, dimethylamino group, diethylamino group, benzyloxycarbonyl group, tert-butoxycarbonyl group, (isopropylisobutyl)methyl group, (bisisopropyl)methyl group, (isopropylcyclohexyl)methyl group, (isobutylcyclohexyl)methyl group, (cyclopentylcyclohexyl)methyl group, (1,2,5-trimethyl)cyclohexyl group, 1-phenylcyclohexyl group, 1-naphthylcyclohexyl group, triphenylmethyl group, 1-naphthylisobutyl group, 1-phenylisopropyl group, 1-cyclohexyl group.

[0066] Example of PCCP ligand synthesis: At room temperature, copper iodide and cesium carbonate in a molar ratio of 1:2 were placed in a reaction flask, and 20 mL of DMF and a certain amount of alkylphosphine and diphenylphosphine were added sequentially. The resulting mixture was stirred at 90 °C for 6 h. The solvent was removed under vacuum, and the product was purified by silica gel column chromatography with a yield of 56%.

[0067] Example of a reaction in the Cr-PCCP system: Ethylene oligomerization was carried out in a 120 mL reactor containing PCCP ligands. The reactor was dried at 120 °C for 3 h under vacuum and then cooled to the desired reaction temperature. The Cr pre-catalyst was weighed into a container under nitrogen atmosphere, and methylcyclohexane was added before transferring the co-catalyst to the container. The resulting mixture was stirred for 1 minute and then immediately transferred to the reactor. The reactor was then immediately pressurized. After the specified reaction time, the ethylene feed was shut off, the system was cooled at 0 °C, the pressure was reduced, and 30 mL of 10% HCl was added for quenching to stop the reaction. A small sample of the supernatant was filtered through a Celite layer, with nonane as an internal standard. The reaction products were analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results included a coselectivity of 0.937 for 1-hexene and 1-octene, and a catalyst activity of 157346 g / (gCr h).

[0068] Similarly, 13 PCCP ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0069] Example 3

[0070] The PNSiP type ligand in the ethylene trimer and tetramer system Cr-PNSiP has the following structural formula (3):

[0071]

[0072] The selection range of R1, R2, and R3 groups is consistent with that in formula (2); R4 group is selected from one or more of the following groups: hydrogen group, methyl group, ethyl group, propyl group, butyl group, cyclohexyl group, phenyl group, allyl group, tolyl group, formyl group, acetyl group, benzoyl group, nitro group, nitrosyl group, fluorine group, bromo group, iodo group, chloro group, amino group, dimethylamino group, diethylamino group, benzyloxycarbonyl group, tert-butyloxycarbonyl group, etc.

[0073] Example of PNSiP ligand synthesis: At room temperature, a mixture of n-butyllithium, 2,6-diisopropylaniline, and n-hexane was stirred overnight, followed by the addition of a certain amount of diphenylphosphine chloride. The precipitated lithium chloride was filtered, and the pale yellow solution was concentrated to obtain a yellow residue. The residue was recrystallized in n-hexane to give N-(2,6-diisopropylpentyl)-1-dipentylphosphine. The product was then obtained by vacuum distillation at 140–150 °C and 10 mmHg, with a yield of 65%.

[0074] Example of a reaction in the Cr-PNSiP system: Ethylene oligomerization was carried out in a 140 mL transparent glass reactor containing PNSiP ligands. Before use, the reactor was heated in a 105°C high-temperature drying oven for 2 hours. The reactor was first evacuated, then purged with high-purity nitrogen; this process was repeated three times. The above steps were then repeated with ethylene. Next, a methylcyclohexane solution of the catalyst precursor and the co-catalyst EADC were introduced into the reactor, followed by ethylene at 1.0 MPa. The reaction was allowed to run for the required time (typically 30 minutes), after which the ethylene feed was shut off. The oligomerization product was cooled in an ice-cold ethanol bath and quenched by the slow addition of 2 mL of acidic ethanol. After venting, the reaction product was analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results included a coselectivity of 0.904 for 1-hexene and 1-octene, and a catalyst activity of 435900 g / (gCr h).

[0075] Similarly, 15 PNSiP type ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0076] Example 4

[0077] The ethylene trimer and tetramer system Cr-PNSiNP has the following structural formula (4):

[0078]

[0079] The selection range of R1, R2, R3, and R4 groups is consistent with that in formula (3); R5 group is selected from one or more of the following groups: hydrogen group, methyl group, ethyl group, propyl group, butyl group, cyclohexyl group, phenyl group, allyl group, tolyl group, formyl group, acetyl group, benzoyl group, nitro group, nitrosyl group, fluorine group, bromo group, iodo group, chloro group, amino group, dimethylamino group, diethylamino group, benzyloxycarbonyl group, tert-butoxycarbonyl group, (isopropylisobutyl)methyl group, (bisisopropyl)methyl group, (isopropylcyclohexyl)methyl group, (isobutylcyclohexyl)methyl group, (cyclopentylcyclohexyl)methyl group, (1,2,5-trimethyl)cyclohexyl group, 1-phenylcyclohexyl group, 1-naphthylcyclohexyl group, triphenylmethyl group, 1-naphthylisobutyl group, 1-phenylisopropyl group, 1-cyclohexyl group.

[0080] Example of PNSiNP ligand synthesis: At -80℃, 4 mol of pyrrole was dissolved in 360 mL of dichloromethane, and the mixture was cooled. Stirring was started, and 3.5 mol of n-butyllithium was slowly added dropwise to the mixture. After 5 min, 2 mol of methyl tert-butyldichlorosilane and 0.22 mol of trifluoroacetic acid were added. After reacting for 5 h, the mixture was placed at room temperature for 36 h. The insoluble matter in the mixture was filtered off, and the mixture was cooled to -2℃. 5 mol of diphenylphosphine chloride was slowly added dropwise to the filtrate. After reacting for 3 h, the mixture was placed at room temperature for 8 h. The mixture was purified by column chromatography, eluted with tetrahydrofuran, and the solvent was evaporated to obtain a white solid powder, which was the desired product with a yield of 75%.

[0081] Example of a reaction in the Cr-PNSiNP system: The ethylene oligomerization reaction was carried out in a 300 mL autoclave. Purified methylcyclohexane was used as the solvent. Before the reaction, the autoclave was heated to 130°C, evacuated for 2 hours, and purged with nitrogen three times. After cooling to room temperature, ethylene was purged twice. First, 95 mL of dehydrated and deoxygenated methylcyclohexane and methylaluminoxane (MAO) were added, followed by a methylcyclohexane solution containing chromium acetylacetonate (Cr(acac)3), PNSiNP ligands, and diphenyl sulfide. Once the temperature was stabilized near the reaction temperature, hydrogen gas was sequentially introduced at 0.3 MPa, followed by ethylene gas until the pressure in the autoclave reached 3 MPa to initiate the reaction. The amount of Cr(acac)3 added was 7 μmol Cr, and the molar ratio of Cr(acac)3:PNSiNP ligand:diphenyl sulfide:MAO was 1:2:1:100. The reaction temperature was 45°C, and the reaction time was 30 min. After the reaction was completed, the ethylene inlet valve was closed, and the mixture was rapidly cooled to 10°C with liquid nitrogen. The pressure was then slowly released, and the reactor was unloaded to obtain the ethylene oligomerization product. The reaction product was analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results included: a co-selectivity of 0.891 for 1-hexene and 1-octene, and a catalyst activity of 642104 g / (gCr h).

[0082] Similarly, 15 PNSiNP type ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0083] Example 5

[0084] The ethylene trimer and tetramer system Cr-PNPN has the following structural formula (5):

[0085]

[0086] The selection range of R1, R2, R3, and R4 groups is consistent with that in formula (3).

[0087] Example of PNPN ligand synthesis: 3 mmol of bis(chlorophosphonyl)amine dissolved in toluene was slowly transferred to a mixture of 8 mmol of a suitable secondary or primary amine, 6 mmol of NEt3, and toluene. The solution was stirred at 40 °C for 24 h, and then became turbid. After evaporating all volatile compounds, the residue was absorbed in hot n-hexane and filtered. The solvent evaporated to produce a colorless oil or solid. If desired, the product could be recrystallized from ethanol, n-pentane, or a mixture of tetrahydrofuran / n-hexane to improve purity. 6 mmol of a suitable secondary amine was lithiated with an equal amount of MeLi in Et2O at 0 °C. The solution was then stirred at room temperature for 6 h, cooled again to 0 °C, treated with an ether solution of 3 mmol of bis(chlorophosphonyl)amine, and stirred again at room temperature for 24 h, with a yield of 83%.

[0088] Example of a reaction in the Cr-PNPN system: Ethylene oligomerization was carried out in a 300 mL autoclave. Isobaric ethylene was added to 80 mL of anhydrous solvent and stirred with a magnetic stirrer. After dissolving the Cr compound and PNPN ligand, an appropriate amount of MMAO-3A solution was added. The solution was immediately transferred to the reactor, and the reaction began. When the maximum ethylene uptake (80 g) was reached, or after a predetermined time (60 min), the ethylene inlet valve was closed, the mixture was cooled to room temperature, the pressure was reduced, and the reactor was opened to stop the reaction. The reaction product was analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results included: a coselectivity of 0.921 for 1-hexene and 1-octene, and a catalyst activity of 235857 g / (gCr h).

[0089] Similarly, 12 PNPN type ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0090] Example 6

[0091] The ethylene trimer and tetramer system Cr-(2-C5H4N)2NR, in which the structural formula of the (2-C5H4N)2NR type ligand is shown in the following formula (6):

[0092]

[0093] The selection range of R1, R2, and R3 groups is consistent with that in formula (2); R2' group is selected from one or more of the following groups: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitrosyl, fluoro, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butyloxycarbonyl; R3' group is selected from the following groups: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl It contains one or more of the following groups: methyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, 1-cyclohexyl.

[0094] Example of synthesis of (2-C5H4N)2NR ligand: A predetermined amount of di(2-pyridyl)methyl ketone (L) was dissolved in 20 mL of toluene, a certain amount of 3,5-dimethylaniline was added, and a drop of glacial acetic acid was added as a catalyst. Then, an appropriate amount of molecular sieve was added to remove water, and the mixture was heated under reflux for 3 days. After filtration, a yellow liquid was obtained. The solvent was then removed by vacuum evaporation, and the product was recrystallized with ethanol to obtain a yellow solid. The yield was 87%.

[0095] Example of a reaction in the Cr-(2-C5H4N)2NR system: The ethylene oligomerization reaction was carried out in a 300mL transparent high-pressure glass reactor. After heating and evacuating the reactor, it was purged three times with high-purity nitrogen and twice with ethylene. Solvent and co-catalyst MAO were added sequentially at a specific temperature. After stirring for two minutes, (2-C5H4N)2NR ligand and CrCl3(THF)3 were added. Ethylene was then introduced to the predetermined pressure, and the reaction was carried out for 30 minutes to obtain the ethylene oligomerization product. The reaction product was analyzed using a deionized water phase followed by organic phase analysis. The final oligomerization results included: a coselectivity of 0.884 for 1-hexene and 1-octene, and a catalyst activity of 374148 g / (gCr h).

[0096] Similarly, 11 (2-C5H4N)2NR type ligands were synthesized and ethylene oligomerization was carried out to obtain the corresponding experimental oligomerization results.

[0097] Part Two: Establishing a Database of Catalytic Systems

[0098] Example 7

[0099] Relationship model between ligands and results in Cr-PNP in ethylene trimer and tetramer systems

[0100] Starting from Example 1, 150 Cr-PNP catalytic system structures were established using simulation software. Based on conformational search, the low-energy conformational structures of key intermediates and transition states of the Cr-PNP system were obtained. The relationship between the rate-controlling step energy barrier and experimental results was compared and analyzed. Descriptors of the Cr-PNP system (bond lengths, bond angles, and dihedral angles of each skeleton atom in the Cr catalyst and PNP ligand, etc.) were extracted as independent variables (i.e., parameters), and corresponding experimental results (co-selectivity of 1-hexene and 1-octene and the reactivity of the Cr catalyst) were used as dependent variables. A sample library of ethylene selective oligomerization catalytic systems was established, that is, the database was obtained by establishing a correlation between descriptors and calculation results. Specifically, the preparation of this database included preprocessing to remove obvious errors or outliers that deviated from expected values, followed by data normalization and regularization, to serve as sample data for training machine learning models.

[0101] Similarly, corresponding databases are established based on the data from Examples 2 to 6.

[0102] Part Three: Building Machine Learning Models

[0103] Example 8

[0104] Machine learning models in Cr-PNP systems of ethylene trimerization and tetramerization

[0105] The Scikit-Learn Python library was used to set and train regression variables on a database. The dataset was randomly divided into training and validation sets at a ratio of 30 / 70 to 90 / 10. Various machine learning algorithms were employed to train the model on the training set. During training, 5-fold cross-validation was used to determine the optimal combination of hyperparameters for predictive ability on the training set. The model was then applied to the validation set to evaluate its predictive performance and select the best machine learning model. The machine learning algorithms were selected from two or more regression algorithms, including Ridge Regression, Support Vector Machine, Nearest Neighbor, Bayesian algorithm, Decision Tree, Neural Network, Random Forest, Gaussian Process Regression, Lasso Regression, and Elastic Network. The descriptors were evaluated using these machine learning algorithms, and the performance of each algorithm in predicting convergence results was compared. The results showed that the Random Forest algorithm had the lowest root mean square error among all algorithms; therefore, it was selected as the preferred model. In machine learning, Random Forest is a classifier containing multiple decision trees, where the output class is determined by the mode of the classes output by individual trees, and the overall model exhibits high accuracy and generalization performance. During training, 5-20 random samplings are performed, and 10-30 times cross-validation is used in each iteration to prevent local overfitting and determine regression accuracy.

[0106] Similarly, the optimal machine learning model was selected from the data of Examples 2 to 6 respectively.

[0107] Part Four: Determining the Catalytic System Based on Machine Learning Models

[0108] Based on the optimal machine learning model, the 10 sets of experimental results from Example 1 were used as the test set to perform data fitting on the Cr-PNP system of Example 1. The fitting results are as follows: Figure 2 As shown, the linear correlation coefficient R² obtained from the fitting is 0.96. It should be understood that the experimental results in Example 1 are merely an example and not a limitation. This result fully demonstrates the good agreement between the experiment and the calculation, thus proving the feasibility of using machine learning to assist in design. Therefore, the descriptor of the catalytic system can be elucidated based on the target oligomerization results (selectivity and reactivity), thereby determining the catalytic system. This elucidation process includes selecting the catalytic system through cross-validation of calculations and experiments.

[0109] In summary, the machine learning-based ethylene oligomerization design method provided by this invention, applied to ethylene trimerization and tetramerization, shows a high degree of agreement between experimental and target values. It overcomes the shortcomings of traditional ethylene oligomerization reactions, such as complex transition states and intermediate conformations, and complex mechanisms of ligand electronic effects and steric hindrance. It replaces the traditional trial-and-error catalyst design method, providing a new approach to rapidly design the required catalytic system and its ligands, thereby enabling rapid screening and highly accurate prediction of highly selective catalytic systems. Experimental verification shows that the ligands screened by this invention can significantly improve the co-selectivity of 1-hexene and 1-octene, thereby enhancing the reactivity.

[0110] Further optimization of the schemes suggests that each catalyst system, namely Cr-PNP, Cr-PCCP, Cr-PNSiP, Cr-PNSiNP, Cr-PNPN and Cr-(2-C5H4N)2NR, may exhibit either a Cr(I) / Cr(III) catalytic cycle or a Cr(II) / Cr(IV) catalytic cycle.

[0111] Further optimization of the scheme may include an activator or co-catalyst, which is selected from one or more of triethylaluminum, trimethylaluminum, triisopropylaluminum, triisobutylaluminum, triethylaluminum trichloride, triethylaluminum trichloride, diethylaluminum chloride, ethylaluminum dichloride, methylaluminoxane, and modified methylaluminoxane.

[0112] In a further optimized formulation, the solvent for each catalyst system is selected from aromatic hydrocarbons, straight-chain or cyclic aliphatic hydrocarbons, and ethers. More preferably, the solvent is selected from benzene, toluene, ethylbenzene, cumene, xylene, hexane, octane, cyclohexane, methylcyclopentane, hexene, hepten, diethyl ether, and tetrahydrofuran.

[0113] Further optimization of the scheme: the molar ratio of ligand to chromium in each catalyst system is 0.01-100 (preferably 0.8-3.0), the pressure of the oligomerization reaction is 0.1-20 MPa (preferably 2.0-6.5 MPa), the oligomerization reaction temperature is 30-120℃, and the oligomerization reaction time is 30 min to 8 h.

[0114] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for molecular design of ethylene oligomerization catalysts based on machine learning, characterized in that, include: Extract the descriptors of the catalytic system as independent variables, and the experimental or calculated oligomerization results as dependent variables to establish a database of the catalytic system; By learning from the database, a model relationship is established between the descriptors and the convergence results of the experiments or calculations, thereby obtaining a machine learning model; Based on the machine learning model, the target catalytic system and its ligands are predicted based on the target aggregation results. The catalytic system includes a Cr catalyst and ligands; The ligands include PNP ligands, PCCP ligands, PNSiP ligands, PNSiNP ligands, PNPN ligands, or (2-C5H4N)2NR ligands; R is hydrogen or an organic group, including R1 group, R2 group, R3 group, R4 group, R5 group, R2' group or R3' group; The descriptor includes the bond length, bond angle, and dihedral angle of each skeletal atom of the Cr catalyst and ligand; The oligomerization results are the coselectivity of 1-hexene and 1-octene and the reactivity of the Cr catalyst.

2. The method for molecular design of ethylene oligomerization catalysts based on machine learning according to claim 1, characterized in that, The process of extracting descriptors of the catalytic system as independent variables and experimental or computational oligomerization results as dependent variables to establish a database of the catalytic system includes: A catalytic system was established using simulation software. The low-energy conformational structures of the key intermediates and transition states of the catalytic system were obtained based on conformational search. The relationship between the energy barrier of the rate-controlling step and the experimental results was compared and analyzed to extract the descriptor of the catalytic system. Using the descriptor of the catalytic system as the independent variable and the corresponding experimental results as the dependent variable, the relationship between the descriptor and the corresponding experimental results was obtained. Based on the relationship, a database of the catalytic system was constructed.

3. The method for molecular design of ethylene oligomerization catalysts based on machine learning according to claim 1, characterized in that, After establishing the database of the catalytic system, the process also includes preprocessing the database. The preprocessing process includes removing obvious errors or outliers from the data, and then normalizing and regularizing the data to serve as sample data for training the machine learning model.

4. The method for molecular design of ethylene oligomerization catalysts based on machine learning according to claim 1, characterized in that, The structural formula of the PNP ligand is: The PCCP ligand structure is as follows: The structural formula of the PNSiP ligand is: The structural formula of the PNSiNP ligand is: The structural formula of the PNPN ligand is: The structural formula of the (2-C5H4N)2NR ligand is: Wherein, the ligand structural formula is: The R1 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitrosyl, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl. The R2 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl. The R3 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl. The R4 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl. The R5 group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl. The R2' group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, and tert-butyloxycarbonyl. The R3' group is selected from one or more of the following: hydrogen, methyl, ethyl, propyl, butyl, cyclohexyl, phenyl, allyl, tolyl, formyl, acetyl, benzoyl, nitro, nitroso, fluorine, bromo, iodo, chloro, amino, dimethylamino, diethylamino, benzyloxycarbonyl, tert-butoxycarbonyl, (isopropylisobutyl)methyl, (bisisopropyl)methyl, (isopropylcyclohexyl)methyl, (isobutylcyclohexyl)methyl, (cyclopentylcyclohexyl)methyl, (1,2,5-trimethyl)cyclohexyl, 1-phenylcyclohexyl, 1-naphthylcyclohexyl, triphenylmethyl, 1-naphthylisobutyl, 1-phenylisopropyl, and 1-cyclohexyl.

5. The method for molecular design of ethylene oligomerization catalysts based on machine learning according to claim 1, characterized in that, The process of learning and establishing a model relationship between the descriptors and the convergence results of experiments or calculations through the database to obtain a machine learning model includes the following steps: Based on the database settings and training regression variables, the dataset is randomly divided into training and validation sets in a ratio of 30 / 70 to 90 / 10. The training set is then trained using a machine learning algorithm, and the hyperparameter combination with the best predictive performance in the training set is obtained through cross-validation. An initial machine learning model is then constructed based on the hyperparameter combination. The data in the validation set is input into the initial machine learning model for validation to obtain the model prediction performance results; The optimal machine learning model is selected based on the predicted performance results of the model, and the machine learning model is obtained.

6. The method for molecular design of ethylene oligomerization catalysts based on machine learning according to claim 1, characterized in that, The machine learning algorithm is selected from two or more of the following regression algorithms: Ridge Regression, Support Vector Machine, Nearest Neighbor Algorithm, Bayesian Algorithm, Decision Tree Algorithm, Neural Network Algorithm, Random Forest Algorithm, Gaussian Process Regression Algorithm, Lasso Algorithm, Elastic Network Algorithm, etc.

Images