An artificial intelligence-based system for replacing specific solvents and raw materials in industrial processes.

Abstract

This invention relates to a system and method for replacing specific solvents and raw materials used in industrial processes with eutectic solvents and mixtures that satisfy specific properties, using artificial intelligence. The system is trained and constantly updated using experimental formation results obtained in the laboratory. The platform can determine whether entirely new systems can be formed and predict some of their physical properties. This system is designed to be applied to industrial processes in which specific solvents and raw materials are used, and identifies eutectic solvents that satisfy or exceed the properties required to replace the specific solvent / raw material. Unlike process-based methods, this method is not applied to a specific process, but rather to the process in which the specific solvent / raw material being replaced is used. This invention provides an effective method for reducing the use of specific solvents in industrial processes and promoting the use of environmentally friendly eutectic solvents, using artificial intelligence.

Description

Technical Field

[0001] This application claims priority under 35 U.S.C. § 119(e) to U.S. Non-Provisional Application No. 18 / 631,100, filed Apr. 10, 2024, and U.S. Provisional Application No. 63 / 458,674, filed Apr. 12, 2023, the entire contents of which are incorporated herein by reference.

[0002] This document relates to systems and methods for replacing specific solvents and raw materials used in industrial processes with eutectic solvents and mixtures that satisfy specific properties using artificial intelligence. The system is trained and continuously updated using experimental formation results obtained in the laboratory. The platform can determine whether a completely new system can be formed and predict some of its physical properties. This system is designed to be applied to industrial processes in which specific solvents / raw materials are used, and to identify eutectic solvents and their mixtures that meet or exceed the properties required to replace a specific solvent. Unlike process-based approaches, this method is not applied to a specific process, but rather to the process in which the specific solvent / raw material being replaced is used. This document provides an effective approach for reducing the use of specific solvents and promoting the use of environmentally friendly eutectic solvents in industrial processes using artificial intelligence.

Background Art

[0003] Integrating artificial intelligence (AI) with other technologies should provide many advantages beyond what can be achieved by the technology alone. These advantages include risk reduction, constant and continuous availability, reduction of human error and unbiased decision-making, assistance by digital computers, improved accuracy of repetitive tasks, rapid decision-making, and pattern recognition leading to new and improved results.

[0004] One such use of AI can be seen in U.S. Patent No. 11,1644,478, which seeks to achieve many of these goals by using AI in combination with food. The AI ​​in this patent uses a predictive model that can be trained using the characteristics of raw materials to match the characteristics of a given target food.

[0005] In the field of chemistry, the introduction of AI should ideally be designed to achieve many of these benefits. Because many functions in chemistry are inherently dangerous, the use of AI should reduce risk by allowing computers to perform some functions previously done by individuals, and potentially reduce risk by making decisions more rational. AI can also mitigate risk by scanning databases of chemical reactions, compounds, and / or solvents for known hazards. While AI is expensive to set up, once an AI system is implemented, it can reduce costs by enabling faster and more rational decision-making and reducing errors caused by human error.

[0006] To date, the combination of chemistry, particularly the use of deep eutectic solvents and AI, has been largely unexplored. Deep eutectic solvents, or DES, are solutions of Lewis or Brønsted acids and bases that form eutectic mixtures. Deep eutectic solvents are highly modifiable by altering the structure of their components or by changing the relative ratios of the various components in the mixture. Because they are complex systems with highly variable properties, they have diverse potential applications, including use in catalysis, separation techniques, and electrochemical processes. The parent components of deep eutectic solvents tend to be involved in complex hydrogen bonding networks or hydrophobic interactions, meaning that the mixture tends to have a significantly lower freezing point compared to the parent compounds / components in the mixture. While individual components in the mixture may be solid at room temperature and atmospheric pressure, when mixed together at room temperature and atmospheric pressure, the mixture can become a liquid with a significantly lower freezing point (e.g., -80°C).

[0007] NADES are eutectic mixtures formed by combinations of natural compounds in specific molar ratios. Deep eutectic solvents (DES) are a newly discovered class of solvent mixtures characterized by a significant melting point depression compared to their corresponding neat components. NADES materials are promising for many applications as inexpensive solvents exhibiting numerous tunable physicochemical properties. Because the microscopic mechanisms governing the structure-property relationships of this class of solvents are generally poorly understood, these solvent systems tend to have unpredictable properties. Complex hydrogen bonding and / or hydrophobic interactions have been proposed as the root cause of their melting point depression and some argue that these are attributes explaining their physicochemical properties. By adjusting the relative amounts of various components, properties can be obtained that cannot be obtained with compositions containing the same components but with different relative amounts. However, as NADES-related systems are newly discovered and explored, it is essential to not only understand these supramolecular NADES networks but also to study the systems as dynamic entities using both simulations and experiments to discover their properties.

[0008] The term "eutectic" was first coined in 1884 by the British chemist and physicist Frederick Guthrie. First-generation eutectic solvents were based on mixtures of quaternary ammonium salts with hydrogen bond donors such as amines and / or carboxylic acids. Natural deep eutectic solvents (NADES) are biologically based deep eutectic solvents composed of two or more compounds, generally primary metabolites of plant origin, i.e., organic acids, sugars, alcohols, amines, and amino acids. Water may also be present as a component, as part of the solvent.

[0009] Since Frederick Guthrie coined the term "eutectic," much of the research on eutectic solvents has involved solvent mixtures in which at least one component is a metal-based solvent. However, the leaching of metals from such solvent systems has shown many drawbacks, associated with metal leaching and related health, environmental, and safety issues. Therefore, recent interest has focused on eutectic systems that do not contain metals.

[0010] While deep eutectic solvents / solutions have proven useful in themselves for substituting for and / or complementing non-eutectic solutions / solvents, a system that enables the use of AI in relation to industrial chemical processes and / or in conjunction with the use of even deeper eutectic solvent systems would be useful. [Prior art documents] [Patent Documents]

[0011] [Patent Document 1] U.S. Patent No. 11,1644,478 [Overview of the project]

[0012] The present invention relates to a system and method for replacing specific solvents and raw materials used in several industrial processes with eutectic solvents that satisfy specific properties using artificial intelligence. In one embodiment, the system is trained and / or constantly updated using experimental formation results obtained in the laboratory. In a variation, the platform can determine whether an entirely new system can be formed and predict some, many, and / or all of its physical properties. In one embodiment, the system is designed to be applied to an industrial process in which a specific solvent is used and identifies a eutectic solvent that satisfies (or potentially surpasses) the properties required to replace the specific solvent. In one embodiment, unlike process-based methods, the method does not have to be applied to a specific process and can be applied to processes in which the specific solvent / raw material being replaced is used. In one embodiment, the present invention alternatively and / or additionally provides an effective method for reducing the use of a specific solvent in an industrial process using artificial intelligence and promoting the use of environmentally friendly eutectic solvents. [Brief explanation of the drawing]

[0013] [Figure 1] This is a diagram illustrating a typical process of applying AI to predict new solvents. [Figure 2] This flowchart provides a brief explanation of how to apply AI to predicting new eutectic systems and their characteristics. [Figure 3] This is a diagram showing how to obtain a fingerprint. [Modes for carrying out the invention]

[0014] The present invention relates to a system and method for replacing specific solvents and raw materials used in industrial processes with eutectic solvents that satisfy specific properties, using artificial intelligence. In one embodiment, the system is trained and constantly updated using experimental formation results obtained in the laboratory. In a variation, the platform can determine whether an entirely new system can be formed and predict some of its physical properties. In a variation, the system is designed to be applied to industrial processes in which a specific solvent is used, and the system can identify eutectic solvents that satisfy or exceed the properties required to replace a specific solvent or raw material. In one embodiment, unlike process-based methods, the method is applied to the process in which the specific solvent being replaced is used, rather than to a specific process. In a variation, the present invention provides an effective method for reducing the use of a specific solvent in industrial processes using artificial intelligence and promoting the use of environmentally friendly eutectic solvents.

[0015] In one embodiment, the AI ​​system of the present invention can use any of a plurality of learning algorithms. In a variation, the learning algorithm (or predictive model) may be one or more of the following: linear regression algorithm, logistic regression algorithm, decision tree algorithm, SVM algorithm, naive Bayes algorithm, KNN algorithm, K-means algorithm, random forest algorithm, support vector machine algorithm, gradient boosting algorithm, DBSCAN algorithm, dimensionality reduction algorithm, gradient boosting algorithm, and / or AdaBoosting algorithm, or a combination thereof. In a variation, the model may include supervised learning, unsupervised learning, semi-supervised learning, deep learning algorithm, reinforcement learning, regression methods, instance-based methods, decision tree learning methods, Bayesian methods, kernel methods, clustering methods, association rule learning algorithms, artificial neural network models, dimensionality reduction methods, ensemble methods, and / or other suitable AI techniques. In one embodiment, the algorithm used is the random forest algorithm. The random forest algorithm is commonly used in classification and regression problems. This algorithm constructs decision trees and makes decisions based on the class selected by the most trees.

[0016] Examples: Types of algorithms: Initially, a decision tree-based random forest model was used (via the scikit-learn library). This algorithm was chosen because the initial amount of data was relatively small, and the model tends to be simpler and more efficient than other methods (such as deep learning methods), thus yielding good results in most cases. This is particularly effective when the AI ​​model is combined with chemical processes, specifically when predictions about solvents are essential. Laboratory validation of the predictions was performed using this model, and good results were obtained. In one example, the model gave 42 out of 52 correct predictions.

[0017] Subsequently, as the amount of experimental data increased, a neural network model (deep learning) was evaluated using the TensorFlow library. This algorithm was chosen to continue because it is generally more efficient for more complex tasks, even if the prediction methodology is less interpretable, but it requires more data for training. The results were compared, and this method gave even better results than the random forest (on the same test set using different data than that used to train the model). The neural network model gave an accuracy of 91%, compared to an accuracy of 85% for the random forest model.

[0018] Conversion to parameters used by databases and algorithms: In one embodiment, the model obtains information about the compounds in each mixture through molecular fingerprints (vectorized representations generated by cheminformatics tools that enable virtual screening and mapping of chemical spaces). To this end, the structure of each compound forming the mixture is first represented in SMILES (Simplified Molecular Input Line Entry System) format (a text-based molecular structure recognized by a chemical structure processing program). The SMILES data is converted into fingerprints using the RDkit program. The latter consists of vectors representing the absence (represented by 0) or presence (corresponding to a 1 in the vector) of each substructure constituting each molecule.

[0019] To represent a mixture in the model's training and test data, the corresponding fingerprints of each constituent molecule are multiplied by their percentages, then added together to reduce bias due to the order of the components, and a dataset scaled between 0 and 1 is obtained. Thus, if a compound is present in a mixture, it is weighted using its respective percentage.

[0020] Columns (features) for training the model are weighted equally for different predictions (stability and / or physical properties).

[0021] The database contains combinations of about 5000 compounds. In addition to information on the stability of eutectic mixtures, there is information on various physicochemical properties of eutectic mixtures, including but not limited to properties such as pH, viscosity, density, conductivity, polarity, refractive index, color, odor, antioxidant capacity, and antibacterial capacity.

[0022] In the formation prediction model (which has to predict whether the mixture is stable or not), the problem of imbalance between positive and negative samples was faced, and the first class (i.e., positive) was overrepresented. In fact, also in this example, this problem is interpreted as the majority of the predictions made with new combinations being positive for stability.

[0023] To address this problem, it was decided to use an unsupervised learning approach. First, a base model was generated that included only combinations of compounds that were highly likely to become unstable (judged by advanced knowledge of the chemistry of eutectic solvents), combined with experimental data and literature data. This model was used to classify data obtained from random combinations of compounds. Mixtures labeled as negative for stability were added to the final dataset to expand the negative region used to train a deterministic classifier. The amount of enhanced data was defined according to the accuracy of a test set (after hyperparameter optimization using the accuracy of the validation set) that included only experimentally verified data. By making such modifications, the problem of overrepresentation of positive weighting was significantly reduced, and about 5% of the random mixtures were classified as positive. This result requires experimental verification as to whether any of the 5% positive correspond to true negatives.

[0024] In the physical property prediction model, since there is little available experimental data (the dataset is limited to combinations that form stable mixtures), it was decided to expand the database with predicted values calculated using mixture design prediction. Using a small set of experimental data, mathematical formulas are adjusted according to the components constituting the mixture and their respective molar ratios, and the properties are predicted by modifying their coefficients.

[0025] On the other hand, when each compound in the mixture has a wide variation in its respective properties such as viscosity, the conversion was performed using a simplified scale defined in advance at predefined intervals. To limit the number of responses and the operation of the algorithm, the compounds were classified into groups of data corresponding to ranges. This was done for many physicochemical properties, including but not limited to pH.

[0026] Furthermore, in order to obtain new combination possibilities and compounds that can form new stable mixtures, natural compound databases such as Pubchem and Lotus were used. From these databases, compounds having a certain similarity to compounds known to form eutectic mixtures (by the Tanimoto Similarity Index) were evaluated. The Tanimoto Similarity Index compares the fingerprints of two molecules according to their common characteristics and gives a value between 0 and 1 according to their similarity. Compounds with a Tanimoto Similarity Index value greater than 0.6 were used.

[0027] Training To train the model, two subsets were separated from the initial dataset: a validation set with 100 positives and 100 negatives (external data for the training set to optimize the model) and a test set with 50 positive data and 50 negative data (representing additional unknown data for the model to verify that similar accuracy is achieved for the validation set). All the remaining data is used for training, that is, the model adjusts its parameters using this data.

[0028] Some model parameters, such as tree size in random forests and learning rate in neural networks, are fixed throughout the training process, requiring trial and error to obtain optimal values. For this purpose, the Optuna library was used to perform heuristic search and acquire the best values ​​in terms of accuracy on the validation set.

[0029] The best parameters for the stability prediction model were obtained using the mean classification error (= correctly classified data / total data), while the mean squared error test was used for the physical property predictor.

[0030] In this way, a model was generated with appropriate parameters that minimize the error in the predictions (both in the prediction of stability and physicochemical properties).

[0031] The model ultimately operates from a sequence of fingerprints corresponding to compound combinations multiplied by their molar ratios, and the model returns a value between 0 and 1 for each case, with values ​​greater than 0.5 corresponding to combinations predicted to be stable. In cases of physical property predictions, the result is a value corresponding to the prediction of that property, or, in a simplified approximation, one of the possible intervals of that value.

[0032] Operation description: Various methods have been proposed to make predictions. One method involves starting with a list of components available in the laboratory. The components are mixed, and randomized molar ratios are used. In this example, a large number of real-world examples can be produced. Predictions are then made for various combinations of these. The model assigns values ​​between 0 and 1, with higher values ​​(i.e., closer to 1) indicating greater reliability of the eutectic mixture. This methodology can also be used to determine values ​​for specific physical properties.

[0033] In a modified version, the second method involves introducing a new compound into the application and searching for similar compounds that form a stable mixture. The most similar compound is substituted for this new compound in each mixture, and stability is predicted using the new substitution component and its similarity to the compound it replaces.

[0034] The final method involves using physicochemical properties as input data (whether or not it clarifies the components of the eutectic mixture), allowing the AI ​​to return NADES (natural deep eutectic solvents) that are likely to satisfy these properties. In this case, several predictions (regarding stability and properties) are first made and stored in a database. Finally, a nearest neighbor search is performed to obtain a mixture whose predicted physicochemical characteristics are as close as possible to the desired characteristics.

[0035] In one embodiment, several methods or modes can be used: 1. Randomized Combinations: This method generates randomized combinations of components and molar ratios and predicts their stability as eutectic mixtures. The model returns values ​​between 0 and 1, allowing for the ranking of the most reliable eutectic mixtures. In addition, this method can predict certain physical properties. 2. Substitution of Similar Compounds: In this mode, a new compound is introduced, and the application searches for similar compounds that form a stable eutectic mixture. The most similar compounds are then replaced with the new compound in their respective mixtures, and the improved stability of the mixture is predicted. 3. Physicochemical Properties: In this mode, the application is given specific physicochemical properties (referring to or not referring to specific components of the eutectic mixture), and the application returns NADES that may satisfy those properties / qualities. In variations, this technique can be used in conjunction with solvent replacement techniques.

[0036] In the modified version, a single training run can achieve verification of random combinations.

[0037] verification In one example, 52 random combinations were created from a list of compounds. Stability predictions for these combinations were performed using a model, predicting 17 to be positive and 35 to be negative for formation. These combinations were then evaluated in the laboratory, and the predictions were correct for 8 of the 17 positives. Of the 35 negative predictions, only 2 were stable mixtures. Therefore, overall, 41 out of 52 predictions (approximately 79%) were correct (8 + 33 = 41), and the majority of true positives (8 out of 10) fell within the subset of positive predictions. Thus, by evaluating only true positives (i.e., ignoring false positives) and negatives, 41 out of 45 combinations yielded accurate results.

[0038] In one embodiment, the combination provided accuracy levels higher than approximately 75%, 80%, 85%, or 90%.

[0039] Further evaluation is needed for formation using more complex neural network methods. Furthermore, further evaluation is needed for predicting physicochemical properties by increasing the data matrix based on equations. Preliminary results are promising compared to previous methods (such as those using the random forest algorithm methodology). This method yields better results than random forests (on the same test set using different data than that used to train the model) (complex neural network methods achieve 91% accuracy compared to random forest methods achieving 85% accuracy).

[0040] As a practical example of solvent substitution in industry, we can mention the extraction of lycopene. Lycopene is an antioxidant found in tomatoes, valued for its properties in combating degenerative diseases. Traditionally, lycopene extraction has been carried out using solvents such as hexane and ethyl acetate, but recently, it has been done using supercritical CO2, which consumes a lot of energy. However, by using artificial intelligence and considering the physicochemical properties of the solvent, we have identified a natural eutectic solvent substitute with similar properties that demonstrated the ability to selectively extract lycopene from tomatoes. This advance may represent a more sustainable and economical alternative for future lycopene extraction.

[0041] Tables 1 and 2 show examples of stability prediction using AI. These results are used to retrain the algorithm for learning and to improve its accuracy.

[0042] [Table 1]

[0043] [Table 2]

[0044] Figure 1 shows a generalized process of how AI can be used for any purpose, including obtaining an idealized solvent system (including model validation, testing, and deployment, as well as subsequent model refinement). Figure 2 shows a more specific AI process, which is the process of selecting an idealized solvent system. Note in this figure that testing the solvent system (evaluating the model) generates new data that allows the solvent system to be continuously improved. Figure 3 shows the use of fingerprint data and the vectors multiplied by their respective percentages to arrive at the data used in the AI ​​system.

[0045] In one embodiment, the present invention relates to a system for identifying ideal solvents or raw materials for use in an industrial process, the system comprising combining a eutectic solvent with artificial intelligence. In a variation, the ideal solvent or component is a mixture of two or more solvents / products. In another variation, the ideal solvent is a mixture of three or more solvents. In yet another variation, the ideal solvent further includes water.

[0046] In one embodiment, the artificial intelligence uses an algorithm, which includes one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof.

[0047] In variations, the algorithm is a random forest algorithm or a complex neural network algorithm. In variations, identifying an ideal solvent involves replacing a specific solvent used in an industrial process with a eutectic solvent. In variations, identifying an ideal solvent involves identifying one or more physicochemical properties. In variations, one or more physicochemical properties include one or more elements selected from the group consisting of pH, viscosity, density, polarity, refractive index, color, odor, antioxidant capacity, and antimicrobial capacity.

[0048] In one embodiment of the system, the accuracy of predicting the ideal solvent is at least about 80%, or at least about 85%, or at least about 90%.

[0049] In one embodiment, the ideal solvent is a mixture of two or more solvents, the ideal solvent further containing water, the artificial intelligence uses an algorithm, the algorithm comprising one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof.

[0050] In one embodiment, the present invention relates to a method for identifying an ideal solvent mixture for use in an industrial process, the method comprising inputting data relating to a plurality of solvents into a computer designed to run an artificial intelligence algorithm, the computer including an artificial intelligence algorithm, the artificial intelligence algorithm processing the data and outputting useful information relating to the solvent mixture, the method comprising inputting, running the algorithm to generate useful information, and evaluating the useful information to identify an ideal solvent mixture.

[0051] In one embodiment, an ideal solvent mixture contains at least one eutectic solvent.

[0052] In the modified version, the artificial intelligence algorithm includes one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof. In the modified version, the algorithm includes a random forest algorithm or a complex neural network algorithm.

[0053] In variations of the method, the data include one or more physicochemical properties selected from the group consisting of pH, viscosity, conductivity, density, polarity, refractive index, color, odor, antioxidant capacity, and antimicrobial capacity.

[0054] In a modified version, the method further includes mixing the solvents and randomizing the molar ratio of the solvents. In another modified version, the molar ratio is not randomized.

[0055] In the modified version, the accuracy of the method for identifying the ideal solvent mixture is at least about 80%. In the modified version, the method includes a training step and a validation step.

[0056] In the modified version, the verification step further includes conducting experiments to determine the ideal solvent mixture.

[0057] It should be understood, intended, and within the scope of the invention that all features listed above can be combined with any other features listed above, provided that they are not incompatible with each other. Wherever a range is mentioned, any real number that falls within that range is intended as the endpoint for generating a subrange. In any case, the invention is defined by the following claims.

Claims

1. A system for identifying an ideal solvent for use in an industrial process, the system comprising combining a eutectic solvent with artificial intelligence.

2. The system according to claim 1, wherein the ideal solvent / component is a mixture of two or more components.

3. The system according to claim 2, wherein the ideal solvent further comprises water as a component.

4. The system according to claim 1, wherein the artificial intelligence uses an algorithm, and the algorithm includes one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof.

5. The system according to claim 4, wherein the algorithm is a random forest algorithm or a complex neural network algorithm.

6. The system according to claim 1, wherein identifying an ideal solvent includes replacing a specific solvent or raw material used in an industrial process with a eutectic solvent.

7. The system according to claim 1, wherein the identification of an ideal solvent includes identifying one or more physicochemical properties.

8. The system according to claim 7, wherein the one or more physicochemical properties include one or more elements selected from the group consisting of pH, viscosity, density, conductivity, polarity, refractive index, color, odor, antioxidant capacity, and antibacterial capacity.

9. The system according to claim 5, wherein the accuracy of predicting the ideal solvent is at least about 80%.

10. The system according to claim 9, wherein the accuracy of predicting the ideal solvent is at least about 90%.

11. The system according to claim 1, wherein the ideal solvent is a mixture of two or more solvents, the ideal solvent further contains water as a component, the artificial intelligence uses an algorithm, the algorithm comprises one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof.

12. A method for identifying an ideal solvent mixture for use in an industrial process, the method being: Inputting data relating to multiple solvents into a computer designed to execute an artificial intelligence algorithm, wherein the computer includes the artificial intelligence algorithm, and the artificial intelligence algorithm processes the data to output useful information relating to a component mixture (eutectic solvent), To generate the aforementioned useful information, the algorithm is executed, To identify the ideal solvent mixture, the useful information is evaluated. Methods that include...

13. The method according to claim 12, wherein the ideal solvent mixture comprises at least one eutectic solvent.

14. The method according to claim 13, wherein the artificial intelligence algorithm includes one or more elements selected from the group consisting of linear regression algorithms, logistic regression algorithms, decision tree algorithms, SVM algorithms, naive Bayes algorithms, KNN algorithms, K-means algorithms, random forest algorithms, complex neural network algorithms, support vector machine algorithms, gradient boosting algorithms, DBSCAN algorithms, dimensionality reduction algorithms, gradient boosting algorithms, AdaBoosting algorithms, and combinations thereof.

15. The method according to claim 14, wherein the algorithm includes a random forest algorithm or a complex neural network algorithm.

16. The method according to claim 12, wherein the data includes one or more physicochemical properties selected from the group consisting of pH, viscosity, conductivity, density, polarity, refractive index, color, odor, antioxidant capacity, and antimicrobial capacity.

17. The method according to claim 12, further comprising mixing the solvent and randomizing the molar ratio of the solvent.

18. The method according to claim 12, wherein the accuracy of the method for identifying the ideal solvent mixture is at least about 80%.

19. The method according to claim 12, wherein the method includes a training step and a verification step.

20. The method according to claim 19, wherein the verification step further comprises conducting an experiment to determine the ideal solvent mixture.

Images