Method and device for predicting solubility of carbon dioxide in ionic solution, and electronic device

Abstract

The application provides a method and device for predicting the solubility of carbon dioxide in an ionic solution and an electronic device, and relates to the technical field of carbon capture, utilization and storage. The method comprises: obtaining a target domain data set corresponding to a first ionic solution for which the solubility of carbon dioxide is to be predicted; and inputting the target domain data set into a carbon dioxide solubility prediction model to obtain a predicted value of the solubility of carbon dioxide under a target working condition parameter corresponding to the first ionic solution. The carbon dioxide solubility prediction model is obtained by performing domain adversarial training on a migration learning model that has been constructed based on a source domain data set and the target domain data set. The features learned by the carbon dioxide solubility prediction model are related to the solubility of carbon dioxide and are independent of the type of data domain. The migration of the solubility of carbon dioxide corresponding to a second ionic solution to the first ionic solution is achieved. Therefore, the accurate prediction of the solubility of carbon dioxide can be achieved based on the carbon dioxide solubility prediction model, and the reliability of the prediction result is significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon capture, utilization and storage, and particularly relates to a method and device for predicting solubility of carbon dioxide in ionic solution and electronic equipment. BACKGROUND

[0002] With the intensification of global climate change, reducing greenhouse gas emissions has become an important task of current environmental governance. Carbon dioxide, as one of the main greenhouse gases, its capture and storage (CCS) technology is considered as one of the key ways to alleviate climate change. In the process of carbon dioxide capture, the performance of the absorbent directly affects the capture efficiency and economy. Traditional carbon dioxide absorbents (such as alcohol amine solutions) have the disadvantages of easy volatilization, strong degradation, high energy consumption, etc., while ionic solutions (ILs) have become a very potential alternative absorbent due to their unique physical and chemical properties such as high thermal stability, low volatility, strong designability and high carbon dioxide solubility. In order to screen ionic solutions with high carbon dioxide absorption efficiency, it is necessary to systematically study the carbon dioxide solubility of ionic solutions under different conditions (such as temperature, pressure, ionic solution structure). At present, the acquisition of carbon dioxide solubility mainly depends on experimental measurement, but the experimental method is time-consuming, high-cost and difficult to cover a wide range of ionic solution types and working conditions. Therefore, based on the calculation model, the prediction of carbon dioxide solubility becomes an important means to supplement the experiment.

[0003] In the related art, a model that can predict the carbon dioxide solubility of the same type of ionic solution is trained by using a training data set of ionic solutions with known carbon dioxide solubility. The training samples in the training data set are related data corresponding to the same type of ionic solution. However, when the model is used to predict the carbon dioxide solubility of other types of ionic solutions, the prediction result is unreliable. SUMMARY

[0004] Embodiments of the present application provide a method and device for predicting the solubility of carbon dioxide in ionic solution, and electronic equipment, to improve the problem that the prediction result is unreliable when the model is used to predict the carbon dioxide solubility of other types of ionic solutions in the related art.

[0005] In a first aspect, embodiments of the present application provide a method for predicting the solubility of carbon dioxide in ionic solution, comprising:

[0006] obtaining a target domain data set corresponding to a first ionic solution to be predicted for carbon dioxide solubility, the target domain data set including physicochemical property data and target working condition parameters corresponding to the first ionic solution;

[0007] The target domain dataset is input into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility, and a predicted value of the carbon dioxide solubility of the first ionic solution under the target working condition parameter is obtained. The carbon dioxide solubility prediction model is obtained by performing domain adversarial training on the migration learning model based on the source domain dataset and the target domain dataset. The features learned by the carbon dioxide solubility prediction model are related to the carbon dioxide solubility and are independent of the data domain type. The source domain dataset includes the physicochemical property data of the second ionic solution corresponding to the known carbon dioxide solubility and the carbon dioxide solubility under different working condition parameters. The type of the second ionic solution is different from that of the first ionic solution.

[0008] In a possible implementation, the migration learning model includes a feature extraction module, a domain discrimination module, and a prediction module. The carbon dioxide solubility prediction model is trained by iteratively training the migration learning model until an iteration termination condition is met. In each round of training, a batch of data extracted from the source domain dataset forms a first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to the carbon dioxide solubility. A batch of data extracted from the source domain dataset and a batch of data extracted from the target domain dataset form a second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features independent of the data domain type.

[0009] In a possible implementation, updating the parameters of the feature extraction module and the prediction module based on the first training sample includes: extracting first features related to the carbon dioxide solubility in the first training sample based on the feature extraction module; predicting the carbon dioxide solubility in the first training sample based on the first features using the prediction module to obtain first predicted values corresponding to the first training sample; determining a solubility loss value based on the first predicted values and the carbon dioxide solubility corresponding to the first training sample; and updating the parameters of the feature extraction module and the prediction module by back propagation based on the solubility loss value to optimize the extraction ability of the feature extraction module for the carbon dioxide solubility related features.

[0010] In a possible implementation, determining the solubility loss value based on the first predicted values and the carbon dioxide solubility corresponding to the first training sample includes: determining the mean square error of the first predicted values and the carbon dioxide solubility corresponding to the first training sample; and determining the mean square error as the solubility loss value.

[0011] In a possible implementation, the second training sample contains a data domain type label identifying a data source, and updating parameters of the feature extraction module and the domain discrimination module based on the second training sample includes: extracting, based on the feature extraction module, second features related to the data domain type in the second training sample; predicting, based on the second features, the data domain type by using the domain discrimination module to obtain a second prediction value corresponding to the second training sample; determining a data domain classification loss value based on the second prediction value and a data domain type label corresponding to the second training sample; and updating the parameters of the feature extraction module and the domain discrimination module by back propagation based on the data domain classification loss value, to weaken the extraction ability of the feature extraction module for the features related to the data domain type.

[0012] In a possible implementation, the domain discrimination module contains a gradient reversal layer, and updating the parameters of the feature extraction module and the domain discrimination module by back propagation based on the data domain classification loss value includes: taking the negative of a domain classification loss gradient corresponding to the data domain classification loss value by using the gradient reversal layer to obtain a reverse gradient; feeding the reverse gradient back to the feature extraction module to update extraction parameters of the feature extraction module, so that the feature extraction module extracts features that are difficult to be distinguished by the domain discrimination module; and feeding the domain classification loss gradient back to the domain discrimination module to update domain discrimination parameters corresponding to the domain discrimination module, to improve the domain discrimination ability of the domain discrimination module.

[0013] In a possible implementation, determining the data domain classification loss value based on the second prediction value and the data domain type label corresponding to the second training sample includes: determining a cross-entropy loss value corresponding to the second prediction value and the data domain type label corresponding to the second training sample; and determining the cross-entropy loss value as the data domain classification loss value.

[0014] In a second aspect, an embodiment of the present application provides a device for predicting solubility of carbon dioxide in an ionic solution, comprising:

[0015] An acquisition module is configured to acquire a target domain data set corresponding to a first ionic solution whose solubility of carbon dioxide is to be predicted, the target domain data set containing physicochemical property data and target working condition parameters corresponding to the first ionic solution;

[0016] A processing module is configured to input the target domain data set into a carbon dioxide solubility prediction model to predict the solubility of carbon dioxide, to obtain a prediction value of the solubility of carbon dioxide of the first ionic solution under the target working condition parameters, the carbon dioxide solubility prediction model being obtained by performing domain adversarial training on a pre-constructed transfer learning model based on a source domain data set and the target domain data set, features learned by the carbon dioxide solubility prediction model being related to the solubility of carbon dioxide and being independent of a data domain type, the source domain data set containing physicochemical property data of a second ionic solution whose solubility of carbon dioxide is known and the solubility of carbon dioxide under different working condition parameters, and the type of the second ionic solution being different from that of the first ionic solution.

[0017] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0018] Memory is used to store instructions executed by the computer;

[0019] A processor for executing computer-executable instructions stored in memory to implement the method described in any of the first aspects.

[0020] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method described in any of the first aspects.

[0021] Fifthly, this application provides a computer program product, including a computer program that, when executed, implements the method described in any of the first aspects.

[0022] The carbon dioxide solubility prediction method, apparatus, and electronic device provided in this application acquire a target domain dataset corresponding to a first ionic solution whose carbon dioxide solubility is to be predicted. The target domain dataset includes physicochemical property data and target operating parameters corresponding to the first ionic solution. The target domain dataset is input into a carbon dioxide solubility prediction model to predict the carbon dioxide solubility, thereby obtaining the predicted value of carbon dioxide solubility under the target operating parameters corresponding to the first ionic solution. The carbon dioxide solubility prediction model is obtained by performing domain adversarial training on a pre-constructed transfer learning model based on the source domain dataset and the target domain dataset. The features learned by the carbon dioxide solubility prediction model are related to carbon dioxide solubility and are independent of the data domain type. The source domain dataset includes physicochemical property data and carbon dioxide solubility under different operating parameters corresponding to a second ionic solution with known carbon dioxide solubility. The type of the second ionic solution is different from that of the first ionic solution. In this process, a transfer learning model is constructed by using a transfer learning algorithm, and the transfer learning model is trained by using domain adversarial methods to obtain a trained carbon dioxide solubility prediction model. This allows the carbon dioxide solubility prediction model to learn features that are related to carbon dioxide solubility and independent of the data domain type. This enables the transfer of carbon dioxide solubility from a second ion solution with known carbon dioxide solubility to a first ion solution (i.e., a new type of ion solution). As a result, the carbon dioxide solubility prediction model can be used to accurately predict the carbon dioxide solubility in the first ion solution, significantly improving the reliability of the carbon dioxide solubility prediction results. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0024] Figure 1 A schematic diagram illustrating an application scenario of the method for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application.

[0025] Figure 2 A flowchart illustrating a method for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application;

[0026] Figure 3 A schematic diagram of the structure of a transfer learning model provided for an exemplary embodiment of this application;

[0027] Figure 4 Another flowchart illustrating the method for predicting the solubility of carbon dioxide in ionic solutions provided as an exemplary embodiment of this application;

[0028] Figure 5 A schematic diagram comparing the prediction results of different models provided for exemplary embodiments of this application;

[0029] Figure 6 A schematic diagram of a device for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application;

[0030] Figure 7 A schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application.

[0031] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0032] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0033] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0034] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0035] First, let me explain the terms used in this application:

[0036] Transfer learning is an important branch of machine learning that has received widespread attention in image recognition and natural language processing. Its goal is to transfer knowledge learned on one or more tasks to another to solve different but related problems.

[0037] Domain adaptation is a method in transfer learning, applicable to scenarios with two data domains: a source domain and a target domain. The source domain has a large amount of labeled data, while the target domain has unlabeled data. The source and target domains share common features and categories, but their feature distributions differ. Domain adaptation aims to learn a model using the information-rich source domain data and apply it to the target domain. The most common approach to domain adaptation is to map the data from both the source and target domains to the same feature space, making the feature distributions of the data in that space as similar as possible, so that the information learned from the source domain can be used in the target domain.

[0038] In related technologies, when using a training dataset of ionic solutions with known carbon dioxide solubility to train a carbon dioxide solubility prediction model, the model's learned features and patterns are limited to that type of ionic solution because the training data only covers specific types of ionic solutions. This makes it difficult to encompass the complex influencing factors and variation patterns of carbon dioxide solubility in other types of ionic solutions. Therefore, when the model encounters a new type of ionic solution that differs significantly from the training data, it cannot effectively transfer learned knowledge, thus failing to accurately predict its carbon dioxide solubility, leading to unreliable prediction results. Furthermore, the inventors discovered during their research that many carbon dioxide solubility prediction models have been proposed in recent years. Based on their design principles, these models fall into two other categories: 1) Thermodynamic models, such as PSRK (Predictive Soave-Redlich-Kwong) or statistical correlation fluid theory based on group contributions. However, these methods only consider some influencing factors, resulting in low accuracy and affecting reliability; 2) Quantitative structure-property models. Methods such as the Group Contribution (GC) method (relationships) assume that the carbon dioxide solubility of an ionic solution is equal to the sum of the carbon dioxide solubility contributions of all its groups, and that the contribution of groups to carbon dioxide remains constant in any ionic solution. By establishing a linear relationship between the molecular groups and properties of ionic solutions, this method is widely used to predict the physical and chemical properties of ionic solutions, such as heat capacity, viscosity, and carbon dioxide solubility. However, due to the complexity of ion interactions in ionic solutions, the prediction results may be unreliable.

[0039] Based on the above, this application provides a scheme for predicting the solubility of carbon dioxide in ionic solutions. It obtains a target domain dataset of the first ionic solution to be predicted, and utilizes a source domain dataset (containing data of a second ionic solution with known carbon dioxide solubility) and the target domain dataset. Based on transfer learning and neural network algorithms, a carbon dioxide solubility prediction model is obtained through domain adversarial training. This model can learn features related to carbon dioxide solubility that are independent of the data domain type, thereby transferring the carbon dioxide solubility information of the second ionic solution with known carbon dioxide solubility to the first ionic solution to be predicted. This allows for accurate prediction of the carbon dioxide solubility in the first ionic solution based on the carbon dioxide solubility prediction model, significantly improving the reliability of the carbon dioxide solubility prediction results.

[0040] Figure 1 This is a schematic diagram illustrating an application scenario of the method for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application. (Example:) Figure 1As shown, this application scenario includes a client 11 and a server 12, wherein the number of clients 11 can be at least one. In practical applications, for example, after researchers or other relevant personnel obtain the target domain dataset corresponding to the first ionic solution for which the carbon dioxide solubility is to be predicted, they send the target domain dataset to the server 12 through the client 11; when the server 12 detects the target domain dataset sent by the user through the client 11, it executes the carbon dioxide solubility prediction method in ionic solutions provided in this application to obtain the predicted value of carbon dioxide solubility under the target operating parameters corresponding to the first ionic solution, and sends the predicted value of carbon dioxide solubility under the target operating parameters corresponding to the first ionic solution to the client 11, thereby allowing the relevant personnel to obtain the predicted value of carbon dioxide solubility under the target operating parameters corresponding to the first ionic solution.

[0041] It should be noted that server 12 can also be replaced by a server cluster or other computing devices with a certain computing power. Client 11 can be a mobile phone, computer, laptop, or personal digital assistant (PDA), etc.

[0042] The following is combined with Figure 1 Application scenarios, refer to Figure 2 This paper describes a method for predicting the solubility of carbon dioxide in ionic solutions according to exemplary embodiments of this application. It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited to... Figure 1 The limitations of the application scenarios shown.

[0043] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0044] Figure 2 This is a schematic flowchart of a method for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application. Figure 2 As shown, the method for predicting the solubility of carbon dioxide in ionic solutions in this application includes the following steps:

[0045] S201. Obtain the target domain dataset corresponding to the first ion solution for which the carbon dioxide solubility is to be predicted. The target domain dataset contains the physicochemical property data and target operating condition parameters corresponding to the first ion solution.

[0046] For example, assuming the first ionic solution for which the carbon dioxide solubility to be predicted is [C2mim][Tf2N], then the target domain dataset corresponding to [C2mim][Tf2N] is obtained. The physicochemical properties in the target domain dataset include, but are not limited to, group types, group numbers, surface tension, density, critical temperature, and critical pressure. The target operating condition parameters in the target domain dataset include target temperature and target pressure.

[0047] S202. Input the target domain dataset into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility and obtain the predicted value of the carbon dioxide solubility of the first ion solution under the target operating conditions. The carbon dioxide solubility prediction model is obtained by training the constructed transfer learning model on domain adversarial training based on the source domain dataset and the target domain dataset. The features learned by the carbon dioxide solubility prediction model are related to the carbon dioxide solubility and are independent of the data domain type. The source domain dataset contains the physicochemical property data of the second ion solution with known carbon dioxide solubility and the carbon dioxide solubility under different operating conditions. The type of the second ion solution is different from the type of the first ion solution.

[0048] Domain Adversarial Training (DAT) is a transfer learning technique that aims to enable models to learn features independent of the data domain type, thereby achieving knowledge transfer and generalization between different but related data domains (such as source and target domains).

[0049] For example, suppose the second ionic solution contains three types: [bmmim][Tf2N], [BMP][Tf2N], and [C8mim][TfO]. The physicochemical properties of each ionic solution and the carbon dioxide solubility under different operating conditions constitute the source domain dataset for the second ionic solution. Accordingly, based on the source domain dataset and the target domain dataset, a pre-built transfer learning model is trained adversarially to obtain a trained carbon dioxide solubility prediction model. The features learned by the carbon dioxide solubility prediction model are related to carbon dioxide solubility and are independent of the data domain type. Correspondingly, the target domain dataset is input into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility of the first ionic solution under the target operating conditions.

[0050] Optionally, before training the pre-built transfer learning model against domain adversarial forces, the process further includes preprocessing the source and target domain datasets separately. Preprocessing operations include, but are not limited to, data cleaning and standardization. For example, in the data cleaning stage, the source and target domain datasets are checked and corrected to remove outliers, missing values, and other potential data quality issues. Outliers may originate from errors during data collection or extreme conditions in specific scenarios, while missing values ​​may stem from incomplete data records or certain special circumstances. Effectively identifying and processing these outliers and missing values ​​can significantly improve data quality, laying a solid foundation for subsequent model training. Correspondingly, in the standardization stage, the features in the source and target domain datasets are uniformly scaled to ensure a relatively balanced contribution of each feature to the model. Standardization eliminates the potential impact of scale differences between different features on model performance, allowing the model to focus more on the intrinsic relationships and patterns between features, thereby improving the accuracy and stability of predictions.

[0051] The carbon dioxide solubility prediction method in ionic solutions provided in this application constructs a transfer learning model using a transfer learning algorithm and trains it with domain adversarial methods to obtain a trained carbon dioxide solubility prediction model. This model learns features related to carbon dioxide solubility that are independent of the data domain type, enabling the transfer of carbon dioxide solubility from a second ionic solution with known solubility to a first ionic solution (i.e., a new type of ionic solution). This allows for accurate prediction of carbon dioxide solubility in the first ionic solution based on the carbon dioxide solubility prediction model, significantly improving the reliability of the prediction results. Furthermore, this prediction method fills the current gap in the field of effective prediction of carbon dioxide solubility in new types of ionic solutions, effectively expanding the application scope of computational models and making their application more practical and feasible. The carbon dioxide solubility prediction model obtained in this way can serve as an important tool for finding efficient carbon dioxide absorption solutions, helping researchers quickly obtain predicted carbon dioxide solubility values ​​for new types of ionic solutions under diverse operating conditions, significantly accelerating the design and development process of new types of ionic solutions. This has positive significance for promoting the development of carbon capture, utilization, and storage technologies.

[0052] In some embodiments, the transfer learning model includes a feature extraction module, a domain discrimination module, and a prediction module. The carbon dioxide solubility prediction model is trained as follows: the transfer learning model is iteratively trained until the iteration termination condition is met, wherein each round of training includes: a batch of data extracted from the source domain dataset constitutes a first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample so that the feature extraction module learns features related to carbon dioxide solubility; a batch of data extracted from the source domain dataset and a batch of data extracted from the target domain dataset constitute a second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample so that the feature extraction module learns features independent of the data domain type.

[0053] For example, Figure 3 A schematic diagram of the structure of a transfer learning model provided for an exemplary embodiment of this application. For example... Figure 3 As shown, the transfer learning model 30 includes a feature extraction module 31, a domain discrimination module 32, and a prediction module 33, wherein:

[0054] Feature extraction module 31 is used to learn and extract features related to carbon dioxide solubility from the input sample data;

[0055] Domain discrimination module 32 is used to determine whether the input sample data belongs to the source domain or the target domain, so as to assist the model in learning domain-invariant features;

[0056] The prediction module 33 is used to output the predicted value of carbon dioxide solubility corresponding to the input sample based on the features extracted by the feature extraction module.

[0057] For example, the parameters of the feature extraction module 31, the domain discrimination module 32, and the prediction module 33 are initialized, for example, by random initialization or pre-trained weight initialization; and the iteration termination condition is set, such as setting the maximum number of iterations, the model's performance index reaching a preset threshold, or the loss function value no longer decreasing significantly as the iteration termination condition; further, the transfer learning model is iteratively trained. Each training round includes the following steps: A batch of data extracted from the source domain dataset constitutes the first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to carbon dioxide solubility; a batch of data extracted from the source domain dataset and a batch of data extracted from the target domain dataset constitute the second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features independent of the data domain type; after each training round, it is checked whether the preset iteration termination condition is met; if it is met, training stops, and the parameters of the current transfer learning model are saved as the final parameters of the carbon dioxide solubility prediction model, and the carbon dioxide solubility prediction model is output; this carbon dioxide solubility prediction model is used to predict the carbon dioxide solubility of samples in the target domain dataset; if it is not met, training continues until the preset iteration termination condition is met.

[0058] In some embodiments, updating the parameters of the feature extraction module and the prediction module based on the first training sample includes: extracting a first feature related to carbon dioxide solubility from the first training sample based on the feature extraction module; predicting the carbon dioxide solubility in the first training sample using the prediction module based on the first feature to obtain a first predicted value corresponding to the first training sample; determining a solubility loss value based on the first predicted value and the carbon dioxide solubility corresponding to the first training sample; and updating the parameters of the feature extraction module and the prediction module through backpropagation based on the solubility loss value to optimize the feature extraction module's ability to extract features related to carbon dioxide solubility.

[0059] For example, the first training sample is input into the transfer learning model. The feature extraction module receives the input data and extracts features related to carbon dioxide solubility through a fully connected layer to obtain the first feature. For example, the functional form of the fully connected layer in the feature extraction module can be expressed as:

[0060]

[0061]

[0062] in, It is the output, for example, the output of the first feature. f(.) is the Rectified Linear Unit (ReLU) activation function, W is the weight, b is the bias, and x is the input. Additionally, the feature extraction module includes a regularization layer, specifically used to adjust the distribution of activation values ​​in each layer to ensure appropriate breadth. This facilitates rapid learning by the transfer learning model without excessive reliance on initial values, thus mitigating the risk of overfitting.

[0063] Accordingly, the extracted first feature The input is fed into the prediction module, which contains a fully connected layer with one neuron. The function corresponding to this fully connected layer can be expressed as:

[0064]

[0065] in, The output is the predicted value of carbon dioxide solubility. Based on the predicted value of carbon dioxide solubility and the carbon dioxide solubility corresponding to the first training sample, the solubility loss value is determined. Further, based on the solubility loss value, the parameters of the feature extraction module and the prediction module are updated through backpropagation to optimize the feature extraction module's ability to extract features related to carbon dioxide solubility. For example, if temperature and pressure have a significant impact on solubility, after the parameter update, the feature extraction module may focus more on the extraction of these features. Through multiple iterations, the transfer learning model gradually extracts features that are highly related to solubility, such as temperature-pressure interaction features.

[0066] In this embodiment, by calculating the solubility loss value and backpropagating to update the parameters, the transfer learning model can gradually optimize the feature extraction module's ability to extract features related to carbon dioxide solubility, thereby significantly improving the performance of the transfer learning model.

[0067] In some embodiments, determining the solubility loss value based on the carbon dioxide solubility corresponding to the first predicted value and the first training sample includes: determining the mean square error of the carbon dioxide solubility corresponding to the first predicted value and the first training sample; and determining the mean square error as the solubility loss value.

[0068] For example, the solubility loss value satisfies the following formula:

[0069]

[0070] Where n is the total number of samples, This is the actual value of carbon dioxide solubility corresponding to the i-th sample. This is the predicted carbon dioxide solubility value corresponding to the i-th sample. This represents the solubility loss value.

[0071] In this embodiment, by using the mean squared error as the formula for calculating the solubility loss value, the degree of deviation between the predicted value and the true value can be intuitively quantified, providing a clear direction for optimization in model training. In addition, its squared penalty mechanism can effectively suppress abnormal error interference, which helps to enhance the stability of the model under complex data and improve the accuracy and reliability of carbon dioxide solubility prediction.

[0072] Based on the above embodiments, in some embodiments, the second training sample includes a data domain type label that identifies the data source. Updating the parameters of the feature extraction module and the domain discrimination module based on the second training sample includes: extracting a second feature related to the data domain type from the second training sample using the feature extraction module; predicting the data domain type using the domain discrimination module based on the second feature to obtain a second predicted value corresponding to the second training sample; determining a data domain classification loss value based on the second predicted value and the data domain type label corresponding to the second training sample; and updating the parameters of the feature extraction module and the domain discrimination module through backpropagation based on the data domain classification loss value to weaken the feature extraction module's ability to extract data domain type-related features.

[0073] For example, the second training sample contains a data domain type label identifying the source of the data. For instance, each sample from the source domain dataset has a data domain type label of 0, indicating that the sample originates from the source domain dataset; each sample from the target domain dataset has a data domain type label of 1, indicating that the sample originates from the target domain dataset. During model training, each round of the second training sample contains N1 samples from the source domain dataset and N2 samples from the target domain dataset, where N1 = N2. Correspondingly, the second training sample is input into the transfer learning model. The feature extraction module receives the input second training sample and extracts features related to the data domain type through a fully connected layer to obtain the second feature. Accordingly, the second feature The data is input into the domain discrimination module, which outputs a predicted data domain type value corresponding to the second training sample through a fully connected layer containing two neurons (the function form is the same as the carbon dioxide solubility output layer). Based on the predicted data domain type value and the data domain type label corresponding to the second training sample, the data domain classification loss value is determined. Further, based on the data domain classification loss value, the parameters of the feature extraction module and the domain discrimination module are updated through backpropagation. Specifically, during the training process, the adversarial learning objectives of the feature extraction module and the domain discrimination module are dynamically adjusted to guide the feature extraction module to gradually weaken its ability to extract data domain type-related features, while enhancing the sensitivity of the domain discrimination module to data domain differences.

[0074] Among them, the domain label output layer, which is used to output data domain type labels, is a fully connected layer with two neurons, and its function representation is similar to that of the carbon dioxide solubility output layer.

[0075] In this embodiment, by using second training samples labeled with data domain types to update the parameters of the feature extraction module and the domain discrimination module, the model can dynamically remove redundant features strongly related to the data source (such as differences in experimental equipment or sampling bias), while strengthening the extraction ability of task-related features that are common across domains (such as physicochemical factors affecting solubility). This adversarial training mechanism, through the classification pressure of the domain discrimination module, inversely constrains the feature extraction module, enabling the model to improve the generalization performance of the target task (such as solubility prediction) while reducing the sensitivity to the data domain, and further enhancing the stability and prediction accuracy of the model in unseen data domains.

[0076] In some embodiments, the domain discrimination module includes a gradient inversion layer. Based on the data domain classification loss value, the parameters of the feature extraction module and the domain discrimination module are updated through backpropagation. This includes: using the gradient inversion layer to invert the domain classification loss gradient corresponding to the data domain classification loss value to obtain the inverse gradient; propagating the inverse gradient back to the feature extraction module to update the extraction parameters of the feature extraction module so that the feature extraction module can extract features that are difficult for the domain discrimination module to distinguish; and propagating the domain classification loss gradient back to the domain discrimination module to update the corresponding domain discrimination parameters of the domain discrimination module to improve the domain discrimination capability of the domain discrimination module.

[0077] For example, adversarial feature alignment can be achieved by using a gradient inversion layer in the domain discrimination module. Specifically, during the forward propagation phase, the gradient inversion layer does not transform the input data, maintaining lossless feature transfer; while during the backpropagation phase, the gradient inversion layer multiplies the gradient from the domain discrimination module by a negative hyperparameter, thus reversing the gradient direction. Without the gradient inversion layer, as training iterations occur, the feature extraction module gradually learns features that can distinguish the source and target domains (such as device noise, sampling patterns, etc.), making the model highly sensitive to the data domain and difficult to generalize to new data domains. However, with the gradient inversion layer, the gradient direction is reversed, forcing the feature extraction module to learn features that cannot be effectively distinguished by the domain discrimination module. During model training, the feature extraction module learns features related to carbon dioxide solubility while simultaneously learning features unrelated to the data domain type, gradually aligning the feature distributions of the source and target domains. Ultimately, the model can extract features strongly correlated with carbon dioxide solubility but unrelated to the data domain type, thus maintaining stable prediction performance in cross-domain scenarios.

[0078] Correspondingly, the corresponding function can be represented as follows:

[0079]

[0080] in, This is the output during forward propagation; The input is α; α is a hyperparameter; during backpropagation, the gradient I is multiplied by -α to obtain the backpropagation gradient. .

[0081] Correspondingly, the reverse gradient The data is fed back to the feature extraction module to update its parameters, enabling the feature extraction module to extract features that are difficult for the domain discrimination module to distinguish; and the gradient of the original domain classification loss is also fed back. The data is sent back to the domain discrimination module to update its parameters, thereby improving the domain discrimination module's domain discrimination capability.

[0082] In this embodiment, by introducing a gradient reversal layer, the feature extraction module receives the reversed gradient during backpropagation, prompting it to generate features for the confusion domain discriminator, weakening data domain-related features, while the domain discriminator is updated normally to improve its discrimination ability. This dynamic adversarial approach forces the model to gradually strip away data domain-specific features and strengthen the extraction of general features that are independent of the data domain, thereby effectively improving the model's generalization performance on new data domains.

[0083] In some embodiments, determining the data domain classification loss value based on the data domain type label corresponding to the second predicted value and the second training sample includes: determining the cross-entropy loss value corresponding to the data domain type label corresponding to the second predicted value and the second training sample; and determining the cross-entropy loss value as the data domain classification loss value.

[0084] For example, the cross-entropy loss value satisfies the following formula:

[0085]

[0086] Where y is the data field type label, which takes the value 0 or 1; It is the data domain type prediction value, i.e. the second prediction value, which ranges between (0,1). The closer it is to 1, the greater the probability that the corresponding sample originates from the target domain dataset. The closer it is to 0, the greater the probability that the corresponding sample originates from the source domain dataset.

[0087] Figure 4 Another schematic flowchart illustrating the method for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application. Figure 4 As shown, the method for predicting the solubility of carbon dioxide in ionic solutions in this application includes the following steps:

[0088] S401. Obtain the target domain dataset corresponding to the first ion solution for which the carbon dioxide solubility is to be predicted. The target domain dataset contains the physicochemical property data and target operating condition parameters corresponding to the first ion solution.

[0089] S402. Obtain the source domain dataset corresponding to the second ion solution with known carbon dioxide solubility. The source domain dataset contains the physicochemical properties data of the second ion solution and the carbon dioxide solubility under different operating conditions.

[0090] The type of the second ion solution is different from that of the first ion solution.

[0091] S403. Conduct domain adversarial training on the constructed transfer learning model based on the source domain dataset and the target domain dataset.

[0092] For example, the transfer learning model is iteratively trained, wherein each round of training includes: a batch of data extracted from the source domain dataset constitutes a first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to carbon dioxide solubility; a batch of data extracted from the source domain dataset and a batch of data extracted from the target domain dataset constitute a second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features independent of the data domain type.

[0093] S404. Determine whether the iteration termination condition is met.

[0094] The iteration termination conditions include, but are not limited to, reaching the maximum number of iterations, the model's performance metrics reaching a preset threshold, or the loss function value no longer decreasing significantly.

[0095] If so, execute S405;

[0096] If not, execute S403.

[0097] S405. Input the target domain dataset into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility and obtain the predicted value of carbon dioxide solubility under the target operating parameters corresponding to the first ion solution.

[0098] For example, assuming the first ionic solution for which the carbon dioxide solubility to be predicted is [bmmim][Tf2N], its corresponding physicochemical property data and target operating parameters constitute the target domain dataset; 19 other types of ionic solution data, such as [BMP][Tf2N], [C2mim][Tf2N], [C2mim][TfO], and [C3mpy][Tf2N], are used as the second ionic solutions with known carbon dioxide solubility, and the corresponding physicochemical property data of the second ionic solutions and the carbon dioxide solubility under different operating parameters constitute the source domain dataset; based on the source domain dataset and the target domain dataset, the carbon dioxide solubility prediction model proposed in this application is trained, for example as... Figure 5The model is named Ours. To demonstrate the effectiveness of this carbon dioxide solubility prediction model, Bayesian Ridge Regression (BR), Support Vector Regression (SVR), Gradient Boosting Regression (GB), Random Forest Regression (RF), and a Neural Network (NN) model without a transfer learning module were trained using the same data. These models were then used to predict the carbon dioxide solubility corresponding to the target domain dataset. The mean squared error was used as the evaluation metric for each prediction result; a smaller error indicates a more accurate prediction. Figure 5 As shown, the model named Ours has the smallest mean square error, indicating that the prediction results obtained by using the carbon dioxide solubility prediction model in the above embodiments are more accurate and significantly improve the reliability of carbon dioxide solubility prediction results.

[0099] In summary, this application has at least the following advantages:

[0100] First, by constructing a transfer learning model using a transfer learning algorithm and training it with domain adversarial methods, a trained carbon dioxide solubility prediction model is obtained. This model learns features related to carbon dioxide solubility that are independent of the data domain type, enabling the transfer of carbon dioxide solubility from a second ion solution (with known solubility) to a first ion solution (i.e., a new type of ion solution). This allows for accurate prediction of carbon dioxide solubility in the first ion solution, significantly improving the reliability of the prediction results. Furthermore, this prediction method fills the current gap in the field of effective prediction of carbon dioxide solubility in new types of ion solutions, effectively expanding the application scope of computational models and making their application more practical and feasible. The carbon dioxide solubility prediction model obtained in this way can serve as an important tool for finding efficient carbon dioxide absorption solutions, helping researchers quickly obtain predicted carbon dioxide solubility values ​​for new types of ion solutions under diverse operating conditions, significantly accelerating the design and development process of new types of ion solutions. This has positive significance for promoting the development of carbon capture, utilization, and storage technologies.

[0101] Second, by calculating the solubility loss value and backpropagating to update the parameters, the transfer learning model can gradually optimize the feature extraction module's ability to extract features related to carbon dioxide solubility, thereby significantly improving the performance of the transfer learning model. Furthermore, by using the mean squared error as the formula for calculating the solubility loss value, the degree of deviation between the predicted value and the true value can be intuitively quantified, providing a clear direction for model training optimization. In addition, its squared penalty mechanism can effectively suppress abnormal error interference, which helps to enhance the stability of the model under complex data and improve the accuracy and reliability of carbon dioxide solubility prediction.

[0102] Third, by using second training samples with data domain type labels to update the parameters of the feature extraction module and the domain discrimination module, the model can dynamically remove redundant features that are strongly related to the data source (such as differences in experimental equipment or sampling bias), while strengthening the ability to extract task-related features that are common across domains (such as physicochemical factors affecting solubility). This adversarial training mechanism, through the classification pressure of the domain discrimination module, inversely constrains the feature extraction module, enabling the model to improve the generalization performance of the target task (such as solubility prediction) while reducing the sensitivity to the data domain, and further enhancing the stability and prediction accuracy of the model in unseen data domains.

[0103] Fourth, by introducing a gradient reversal layer, during backpropagation, the feature extraction module receives the reversed gradient, prompting it to generate features for the confusion domain discriminator, weakening data domain-related features, while the domain discriminator is updated normally to improve its discriminative ability. This dynamic adversarial approach forces the model to gradually strip away data domain-specific features and strengthen the extraction of general features that are independent of the data domain, thereby effectively improving the model's generalization performance on new data domains.

[0104] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0105] Figure 6 A schematic diagram of a device for predicting the solubility of carbon dioxide in ionic solutions, provided as an exemplary embodiment of this application. Figure 6 As shown, the carbon dioxide solubility prediction device 60 in ionic solutions includes an acquisition module 61 and a processing module 62, wherein:

[0106] The acquisition module 61 is used to acquire the target domain dataset corresponding to the first ion solution whose carbon dioxide solubility is to be predicted. The target domain dataset contains the physicochemical property data and target operating condition parameters corresponding to the first ion solution.

[0107] Processing module 62 is used to input the target domain dataset into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility, and obtain the predicted value of the carbon dioxide solubility of the first ion solution under the target operating parameters. The carbon dioxide solubility prediction model is obtained by training the constructed transfer learning model on domain adversarial training based on the source domain dataset and the target domain dataset. The features learned by the carbon dioxide solubility prediction model are related to the carbon dioxide solubility and are independent of the data domain type. The source domain dataset contains the physicochemical property data of the second ion solution with known carbon dioxide solubility and the carbon dioxide solubility under different operating parameters. The type of the second ion solution is different from the type of the first ion solution.

[0108] In one possible implementation, the transfer learning model includes a feature extraction module, a domain discrimination module, and a prediction module. The carbon dioxide solubility prediction model is trained as follows: the transfer learning model is iteratively trained until the iteration termination condition is met, wherein each round of training includes: a batch of data extracted from the source domain dataset constitutes a first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to carbon dioxide solubility; a batch of data extracted from the source domain dataset and a batch of data extracted from the target domain dataset constitute a second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features independent of the data domain type.

[0109] In one possible implementation, the processing module 62 may be specifically used to: extract a first feature related to carbon dioxide solubility from the first training sample based on the feature extraction module; predict the carbon dioxide solubility in the first training sample using the prediction module based on the first feature to obtain a first predicted value corresponding to the first training sample; determine a solubility loss value based on the first predicted value and the carbon dioxide solubility corresponding to the first training sample; and update the parameters of the feature extraction module and the prediction module through backpropagation based on the solubility loss value to optimize the feature extraction module's ability to extract features related to carbon dioxide solubility.

[0110] In one possible implementation, the processing module 62 can also be used to: determine the mean square error of the carbon dioxide solubility corresponding to the first predicted value and the first training sample; and determine the mean square error as the solubility loss value.

[0111] In one possible implementation, the second training sample contains a data domain type label that identifies the source of the data. The processing module 62 can also be used to: extract a second feature related to the data domain type from the second training sample based on the feature extraction module; predict the data domain type using the domain discrimination module based on the second feature to obtain a second predicted value corresponding to the second training sample; determine a data domain classification loss value based on the second predicted value and the data domain type label corresponding to the second training sample; and update the parameters of the feature extraction module and the domain discrimination module through backpropagation based on the data domain classification loss value to weaken the feature extraction module's ability to extract features related to the data domain type.

[0112] In one possible implementation, the domain discrimination module includes a gradient inversion layer, and the processing module 62 can also be used to: use the gradient inversion layer to invert the domain classification loss gradient corresponding to the data domain classification loss value to obtain the inverse gradient; feed the inverse gradient back to the feature extraction module to update the extraction parameters of the feature extraction module so that the feature extraction module can extract features that are difficult for the domain discrimination module to distinguish; feed the domain classification loss gradient back to the domain discrimination module to update the domain discrimination parameters corresponding to the domain discrimination module so as to improve the domain discrimination capability of the domain discrimination module.

[0113] In one possible implementation, the processing module 62 can also be used to: determine the cross-entropy loss value corresponding to the data domain type label of the second predicted value and the second training sample; and determine the cross-entropy loss value as the data domain classification loss value.

[0114] The carbon dioxide solubility prediction device in ionic solutions provided in this application embodiment can execute the technical solution shown in the above-described carbon dioxide solubility prediction method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0115] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, a processing module can be a separate processing element, or it can be integrated into a chip within the device. Alternatively, it can be stored as program code in the device's memory, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0116] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-On-a-Chip (SOC).

[0117] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Video Discs, DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0118] Figure 7 A schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. For example... Figure 7 As shown, the electronic device 70 in this embodiment includes:

[0119] At least one processor 71; and a memory 72 communicatively connected to said at least one processor;

[0120] The memory 72 stores instructions that can be executed by the at least one processor 71 to cause the electronic device to perform the method as described in any of the above embodiments.

[0121] Alternatively, the memory 72 can be either standalone or integrated with the processor 71.

[0122] The memory 72 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0123] The processor 71 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Specifically, when implementing the model training method or VSP wave field separation method described in the foregoing method embodiments, the electronic device may be, for example, an electronic device with processing capabilities such as a server.

[0124] Optionally, the electronic device may also include a communication interface 73. In specific implementations, if the communication interface 73, memory 72, and processor 71 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0125] Optionally, in a specific implementation, if the communication interface 73, memory 72 and processor 71 are integrated on a single chip, then the communication interface 73, memory 72 and processor 71 can communicate through an internal interface.

[0126] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0127] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed, they are used to implement the method steps as described in the above method embodiments. The specific implementation methods and technical effects are similar and will not be repeated here.

[0128] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0129] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in a device for predicting the solubility of carbon dioxide in ionic solutions.

[0130] This application also provides a computer program product, including a computer program, which, when executed, implements the method steps as described in the above method embodiments. The specific implementation and technical effects are similar and will not be repeated here.

[0131] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0132] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for predicting the solubility of carbon dioxide in ionic solutions, characterized in that, include: Obtain the target domain dataset corresponding to the first ion solution for which the carbon dioxide solubility is to be predicted. The target domain dataset contains the physicochemical property data and target operating condition parameters corresponding to the first ion solution. The physicochemical property data includes the type of functional group, the number of functional groups, surface tension, density, critical temperature and critical pressure. The target operating condition parameters include the target temperature and the target pressure. The target domain dataset is input into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility, thereby obtaining the predicted value of the carbon dioxide solubility of the first ion solution under the target operating conditions. The carbon dioxide solubility prediction model is obtained by training a pre-constructed transfer learning model on domain adversarial training based on the source domain dataset and the target domain dataset. The transfer learning model includes a feature extraction module, a domain discrimination module, and a prediction module. The domain discrimination module includes a gradient inversion layer. The features learned by the carbon dioxide solubility prediction model are related to carbon dioxide solubility and are independent of the data domain type. The source domain dataset includes the physicochemical property data of the second ion solution with known carbon dioxide solubility and the carbon dioxide solubility under different operating conditions. The type of the second ion solution is different from the type of the first ion solution. The carbon dioxide solubility prediction model was trained in the following manner: The transfer learning model is trained iteratively until the iteration termination condition is met, wherein each round of training includes: The batch data extracted from the source domain dataset constitutes the first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to carbon dioxide solubility. The batch data extracted from the source domain dataset and the batch data extracted from the target domain dataset constitute the second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features that are independent of the data domain type.

2. The prediction method according to claim 1, characterized in that, The step of updating the parameters of the feature extraction module and the prediction module based on the first training samples includes: Based on the feature extraction module, a first feature related to carbon dioxide solubility is extracted from the first training sample; Based on the first feature, the prediction module is used to predict the carbon dioxide solubility in the first training sample to obtain the first predicted value corresponding to the first training sample. Based on the carbon dioxide solubility corresponding to the first predicted value and the first training sample, determine the solubility loss value; Based on the solubility loss value, the parameters of the feature extraction module and the prediction module are updated through backpropagation to optimize the feature extraction module's ability to extract carbon dioxide solubility-related features.

3. The prediction method according to claim 2, characterized in that, The step of determining the solubility loss value based on the carbon dioxide solubility corresponding to the first predicted value and the first training sample includes: Determine the mean square error between the first predicted value and the carbon dioxide solubility corresponding to the first training sample; The mean square error is determined to be the solubility loss value.

4. The prediction method according to any one of claims 1 to 3, characterized in that, The second training sample contains data domain type labels that identify the data source. Updating the parameters of the feature extraction module and the domain discrimination module based on the second training sample includes: Based on the feature extraction module, a second feature related to the data domain type is extracted from the second training sample; Based on the second feature, the domain discrimination module is used to predict the data domain type to obtain the second predicted value corresponding to the second training sample; Based on the second predicted value and the data domain type label corresponding to the second training sample, determine the data domain classification loss value; Based on the data domain classification loss value, the parameters of the feature extraction module and the domain discrimination module are updated through backpropagation to weaken the feature extraction module's ability to extract data domain type-related features.

5. The prediction method according to claim 4, characterized in that, The step of updating the parameters of the feature extraction module and the domain discrimination module through backpropagation based on the data domain classification loss value includes: Using the gradient inversion layer, the gradient of the domain classification loss corresponding to the data domain classification loss value is inverted to obtain the inverse gradient; The inverse gradient is fed back to the feature extraction module to update the extraction parameters of the feature extraction module, so that the feature extraction module can extract features that are difficult for the domain discrimination module to distinguish. The gradient of the domain classification loss is backpropagated to the domain discrimination module to update the domain discrimination parameters corresponding to the domain discrimination module, thereby improving the domain discrimination capability of the domain discrimination module.

6. The prediction method according to claim 4, characterized in that, The step of determining the data domain classification loss value based on the second predicted value and the data domain type label corresponding to the second training sample includes: Determine the cross-entropy loss value corresponding to the data domain type label of the second predicted value and the second training sample; The cross-entropy loss value is determined to be the data domain classification loss value.

7. A device for predicting the solubility of carbon dioxide in ionic solutions, characterized in that, include: The acquisition module is used to acquire the target domain dataset corresponding to the first ion solution for which the carbon dioxide solubility is to be predicted. The target domain dataset contains the physicochemical property data and target operating condition parameters corresponding to the first ion solution. The physicochemical property data includes the type of functional groups, the number of functional groups, surface tension, density, critical temperature and critical pressure. The target operating condition parameters include the target temperature and the target pressure. The processing module is used to input the target domain dataset into the carbon dioxide solubility prediction model to predict the carbon dioxide solubility, thereby obtaining the predicted value of the carbon dioxide solubility of the first ion solution under the target operating conditions. The carbon dioxide solubility prediction model is obtained by training a pre-constructed transfer learning model on domain adversarial training based on the source domain dataset and the target domain dataset. The transfer learning model includes a feature extraction module, a domain discrimination module, and a prediction module. The domain discrimination module includes a gradient inversion layer. The features learned by the carbon dioxide solubility prediction model are related to carbon dioxide solubility and are independent of the data domain type. The source domain dataset includes the physicochemical property data of the second ion solution with known carbon dioxide solubility and the carbon dioxide solubility under different operating conditions. The type of the second ion solution is different from the type of the first ion solution. The carbon dioxide solubility prediction model was trained in the following manner: The transfer learning model is trained iteratively until the iteration termination condition is met, wherein each round of training includes: The batch data extracted from the source domain dataset constitutes the first training sample, and the parameters of the feature extraction module and the prediction module are updated based on the first training sample, so that the feature extraction module learns features related to carbon dioxide solubility. The batch data extracted from the source domain dataset and the batch data extracted from the target domain dataset constitute the second training sample, and the parameters of the feature extraction module and the domain discrimination module are updated based on the second training sample, so that the feature extraction module learns features that are independent of the data domain type.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is used to store computer-executed instructions; The processor is configured to execute the computer execution instructions to implement the method for predicting the solubility of carbon dioxide in ionic solutions as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method for predicting the solubility of carbon dioxide in ionic solutions as described in any one of claims 1 to 6.

Images