Cross-water-source micropollutant adsorption performance prediction model and method under data scarcity condition

Abstract

The invention provides a cross-water-source micropollutant adsorption performance prediction model and method under a data scarcity condition, and the model is constructed by the following steps: collecting an adsorption sample and smelly substances, and carrying out adsorption in ultrapure water and source water to obtain adsorption isotherm data so as to generate an ultrapure water adsorption data set and a source water adsorption data set; an ultrapure water adsorption data set is used for developing a base model, and the optimized deep neural network is used as an application algorithm of the base model; and developing a transfer learning model by using the source water adsorption data set, migrating knowledge of the base model to the transfer learning model, finely adjusting the transfer learning model, and only adjusting an output layer and / or a part of hidden layer of the transfer learning model. According to the model, cross-water-source micropollutant adsorption performance prediction can be carried out, fine adjustment is carried out on the transfer learning model only through two data points, the model can reach the R2 value of 0.75, the training requirement for adsorption data of the smelly substances in water of the water source is reduced, and meanwhile high prediction accuracy is kept.

Description

Technical Field

[0001] The present invention relates to the technical field of water source micropollutant adsorption performance, and in particular to a model and method for predicting the adsorption performance of cross-water source micropollutants under data scarcity conditions. Background Art

[0002] Establishing an efficient adsorption prediction model is of great significance to improving the treatment efficiency of micropollutants under changing water quality conditions and reducing the workload of adsorption experiments. In the operation of activated carbon process, it is necessary to estimate the amount of activated carbon added under different water quality conditions, and efficient and accurate adsorption prediction depends on the precise description of three core elements in the modeling process: the molecular characteristics of micropollutants, the surface characteristics of activated carbon, and the component characteristics of natural organic matter (NOM).

[0003] In the early stages of adsorption prediction research, researchers only focused on the prediction of the adsorption performance of micropollutants in raw water when one of the three factors, namely, micropollutant characteristics, micropollutant concentration, and activated carbon type, changed, in order to guide the optimization of activated carbon selection and dosage in the actual production process. Early researchers found that there were large differences in the adsorption performance characterized by isothermal adsorption lines for micropollutants with different initial concentrations in the same raw water, and NOM had a stronger competitive adsorption capacity for trace levels of micropollutants. The Ideal Adsorbed Solution Theory-Equivalent Background Compound Model proposed by Najm et al. can effectively quantify the effect of NOM on the adsorption performance of micropollutants at different concentrations, thereby predicting the adsorption isotherms at different initial concentrations of micropollutants.

[0004] In terms of the effectiveness of predicting the adsorption capacity of different activated carbons for specific pollutants, the pore structure and surface characteristics of activated carbon are the key factors that determine its adsorption performance. Huang et al. analyzed the effects of various activated carbon characteristics on the adsorption performance of odor-causing substances by single-parameter and multi-parameter linear fitting methods, and found that the linear regression model constructed by activated carbon particle size (D40) and micropore volume can effectively predict the adsorption of MIB in water. The characteristics of using linear regression models in adsorption prediction are that the characteristic variables are clear and the contribution is easy to quantify, which can provide a scientific and easy-to-understand theoretical basis for water plants to select activated carbon for odor removal. In order to evaluate the adsorption performance of activated carbon for unknown micropollutants, the researchers used single-parameter and multi-parameter linear fitting methods to estimate the adsorption performance of unknown pollutants based on the structural chemical parameters of the pollutants. Given the diverse structures of organic pollutants, it is crucial to select appropriate descriptors for micropollutants. Among the many types of descriptors, structural descriptors and physicochemical descriptors are the most commonly used. The former, such as the number of atoms and the number of aromatic rings, directly reflects the basic composition of the molecule, while the latter, such as polarity and hydrophobicity, reveals the adsorption mechanism. As the most widely used physicochemical descriptor to date, LSER uses only five parameters, E, S, A, B and V, to fully characterize the structural variability of various organic pollutants. The prediction model constructed based on this is generally highly interpretable.

[0005] Recently, the research focus of adsorption modeling has gradually shifted to the prediction of pollutant adsorption performance under the scenario of synchronous changes in multiple factors. The rapid development and application of machine learning technology in recent years is expected to solve the problem of synchronous prediction of removal efficiency of hundreds of organic pollutants in the adsorption field. Thanks to the emergence of machine learning algorithms, the adsorption prediction model can realize the simultaneous investigation of the influence of adsorbent-adsorbate characteristics on the adsorption performance of activated carbon, analyze the nonlinear relationship between various factors and give reasonable prediction values ​​of the adsorption performance of different micropollutants. Zhang et al. selected a deep neural network algorithm from a variety of machine learning algorithms (neural networks, XGBoost, random forests, etc.) and expanded the effective range of the adsorption prediction model to 165 organic pollutants and 35 activated carbons. The expansion of the adsorption modeling data set is also conducive to the improvement of the model's generalization ability and the guarantee of prediction accuracy. However, at present, the model can only be applied to the adsorption prediction of a single solute adsorption system (under the background of pure water), and lacks the ability to be applied in actual water treatment processes.

[0006] In addition, in the face of the adsorption prediction of odor-causing substances of different concentrations in different water sources, Zietzschmann et al. combined the IAST-EBC model with the NOM characteristic parameters based on exclusion chromatography to achieve the prediction of micropollutant adsorption performance under different water source conditions. In future studies, the cross-water source prediction model will be effectively combined with the pollutant structure-activity relationship model, which is expected to achieve accurate prediction of adsorption efficiency under different micropollutant-activated carbon-water source water quality combinations, and further improve the scope of application and application potential of the adsorption prediction model in water treatment operation management.

[0007] High-quality research data in the field of adsorption are mainly concentrated in single-solute adsorption systems in pure water environments, while the actual application scenarios of adsorption models are mainly in the prediction and control of pollutant adsorption in actual water sources. Water companies are not capable of collecting and accumulating pollutant adsorption data. They can generally only use activated carbon adsorbents used by local water plants to conduct adsorption experiments and limited data collection on common micropollutants in local water sources. Since the source and scale of data sets are usually limited, the development and application of large-scale machine learning adsorption prediction models are also severely restricted.

[0008] The above-mentioned record of background technical knowledge is intended to help ordinary technicians in this field understand the existing technologies that are relatively close to the present invention, and at the same time facilitate the understanding of the inventive concept and technical solution of the present invention. It should be clear that in the absence of clear evidence that the above-mentioned content has been disclosed before the filing date of this patent application, the above-mentioned background technology should not be used to evaluate the novelty of the technical solution of this application. Summary of the invention

[0009] Technical issues In order to solve the above problems, the purpose of the present invention is to provide a model and method for predicting the adsorption performance of micropollutants across water sources under data scarcity conditions, collect adsorption samples, and adsorption data sets of odor-causing substances in ultrapure water and source water to develop a base model and a transfer learning model. The transfer learning model can be fine-tuned with very little data to enable the model to achieve a higher prediction accuracy. Therefore, it is suitable for situations where there is less adsorption data in specific waters, requires less computing resources, a small number of training samples, and the prediction accuracy reaches a practical level.

[0010] Technical Solution In order to achieve the above objectives, the inventors of the present application conducted in-depth research and found that a transfer learning model was developed based on a water source water adsorption dataset, and a base model was pre-trained using an ultrapure water adsorption dataset. The knowledge of the base model was transferred to the transfer learning model and fine-tuned to obtain an optimized transfer learning model. This model can be used to predict the adsorption performance of micropollutants across water sources. By fine-tuning the transfer learning model with only two data points, the model can achieve an R of 0.75.2 The value reduces the training requirements for source water odorant adsorption data while maintaining high prediction accuracy.

[0011] That is, the present invention includes the following aspects.

[0012] (1) Method for constructing a prediction model for the adsorption performance of micropollutants across water sources, including: Step 1, obtaining an adsorption data set: collecting at least one adsorption sample and at least one odor-causing substance, and performing adsorption in ultrapure water and source water to obtain adsorption isotherm data, so as to generate an ultrapure water adsorption data set and a source water adsorption data set; Step 2: Build a base model: Develop a base model using an ultrapure water adsorption dataset, and use an optimized deep neural network as the application algorithm of the base model. The deep neural network includes an input layer, three hidden layers, and an output layer, and a five-fold cross validation is performed. The base model in the present invention is a basic, general, pre-trained model that can be fine-tuned on specific tasks to improve performance. In the present invention, it is used to build a more complex transfer learning model. Step 3. Build a transfer learning model: Develop a transfer learning model with the source water adsorption dataset, pre-train the base model with the ultrapure water adsorption dataset, transfer the knowledge of the base model to the transfer learning model and fine-tune it, and only adjust the output layer and / or part of the hidden layer of the transfer learning model.

[0013] In some specific embodiments of the present invention, in step 1, the adsorption performance is expressed by the solid-liquid partition coefficient logK d To evaluate, as shown in formula (1), Formula (1) Among them, c e Represents different equilibrium concentrations; q e It indicates the amount of chemical substance on the solid surface per unit mass or unit area when adsorption equilibrium is reached; using the solid-liquid partition coefficient logKd to evaluate adsorption performance can better compare the adsorption conditions at different equilibrium concentrations.

[0014] In some specific embodiments of the present invention, in step 1, five Abraham descriptors E, S, A, B, and V are used to describe the characteristics of odor-causing substances, and the total pore volume (Vt) and BET surface area are used as descriptors of the adsorption sample. For the adsorption conditions, c e is the only input feature.

[0015] In some specific implementation schemes of the present invention, in step 2, the three hidden layers have 128, 64 and 32 neurons respectively.

[0016] In some specific embodiments of the present invention, in step 2, in order to minimize data leakage, the data are grouped according to different adsorbents during model development, validation and evaluation.

[0017] In some specific implementation schemes of the present invention, in step 2, the base model evaluation indicators include R² and root mean square error RMSE.

[0018] According to some specific implementation schemes of the present invention, in step three, a small number of samples are used to fine-tune the parameters of the transfer learning model with at least 2 data points; specifically, 2-100 of the more than 600 data points are used to fine-tune the parameters of the transfer learning model, and the remaining data points are used to evaluate the degree of understanding and generalization performance of the transfer learning model on the adsorption behavior of odor-causing substances in water source water.

[0019] In some specific implementation schemes of the present invention, in step three, the transfer learning model effectively inherits the weights and neuron connections learned in the base model training.

[0020] In the scenario of predicting the adsorption of odor-causing substances in source water, the base model and the transfer learning model are constructed using the ultrapure water adsorption dataset and the source water adsorption dataset respectively. When fine-tuning the transfer learning model, the parameters of the base model are inherited and fixed, and only the output layer or part of the hidden layer is adjusted. The pre-trained knowledge is the base model for the prediction of the adsorption of odor-causing substances in pure water. This model contains extensive knowledge about the adsorption mechanism of pollutants, and the transfer learning model can adapt this knowledge to the understanding of the mechanism of adsorption of odor-causing substances in source water. The obtained transfer learning model still shows excellent performance under limited data conditions. With very small sample size for training, the transfer learning model still achieves excellent prediction performance in the adsorption of odor-causing substances in source water. The model can achieve an R of 0.75 by fine-tuning the transfer learning model with only two data points. 2 The value reduces the training requirements for source water odorant adsorption data while maintaining high prediction accuracy.

[0021] (2) A prediction model for the adsorption performance of micropollutants across water sources obtained based on the aforementioned construction method.

[0022] (3) The method for predicting the adsorption performance of micropollutants across water sources under data scarcity conditions is based on the aforementioned prediction model for the adsorption performance of micropollutants across water sources.

[0023] (4) A computer device, comprising a memory, a processor, a communication interface and a communication bus; wherein the memory, the processor and the communication interface communicate with each other via the communication bus; the memory is used to store computer programs; the processor is used to execute the computer programs stored in the memory, and when the processor executes the computer program, it implements the method for constructing a model for predicting the adsorption performance of micropollutants across water sources or the method for predicting the adsorption performance of micropollutants across water sources under data-scarce conditions.

[0024] (5) A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing the aforementioned model for predicting the adsorption performance of micropollutants across water sources or the method for predicting the adsorption performance of micropollutants across water sources under data-scarce conditions.

[0025] Based on the common knowledge in this field, the above-mentioned preferred conditions can be combined with each other to obtain a specific implementation method.

[0026] Beneficial Effects According to the present invention, by collecting adsorption samples and adsorption isotherm data of odor-causing substances in ultrapure water and source water, an ultrapure water adsorption data set and a source water adsorption data set are obtained, a base model is developed with the ultrapure water adsorption data set, a transfer learning model is developed with the source water adsorption data set, the base model is pre-trained with the ultrapure water adsorption data set, the knowledge of the base model is transferred to the transfer learning model and fine-tuned to obtain a tuned transfer learning model, and the adsorption performance of micropollutants across water sources can be predicted based on this model. Based on the transfer learning algorithm, the present invention can realize the adsorption performance prediction of multiple micropollutants and multiple activated carbons when there is less adsorption data in a specific water area, and the transfer learning model can be fine-tuned by only two data points, and the model can achieve an R of 0.75. 2 The model makes full use of the adsorption data of micropollutants in pure water to predict the adsorption of micropollutants in source water under data-scarce scenarios. It requires fewer computing resources and a smaller number of training samples, and its prediction accuracy reaches a practical level, which is suitable for natural water and urban sewage.

[0027] The present invention adopts the above technical solution to achieve the above purpose, which makes up for the shortcomings of the prior art and has reasonable design and convenient operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the above and / or other purposes, features, advantages and examples of the present invention more obvious and easy to understand, the drawings required for use in the specific implementation of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 It represents the comparison between the predicted value and the actual value of the adsorption efficiency Kd of the transfer learning model; Figure 2 Indicates the effect of sample size on the generalization performance of the model; Figure 3 Comparison of training computing resources required for fine-tuning transfer learning models and retraining DNN models; Figure 4 Comparison of training computing resources required to optimize fine-tune the transfer learning model and retrain the DNN model. DETAILED DESCRIPTION

[0030] Those skilled in the art can refer to the content of this article and appropriately replace and / or modify the process parameters to achieve the same. However, it should be particularly noted that all similar replacements and / or modifications are obvious to those skilled in the art and are considered to be included in the present invention. The products and preparation methods described in the present invention have been described through preferred examples. It is obvious that relevant personnel can modify or appropriately change and combine the products and preparation methods described herein without departing from the content, spirit and scope of the present invention to implement and apply the technology of the present invention.

[0031] Unless otherwise defined, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention belongs. The present invention uses the methods and materials described herein; however, other suitable methods and materials known in the art may also be used. The materials, methods, and examples described herein are illustrative only and are not intended to be limiting. All publications, patent applications, patents, provisional applications, database entries, and other references mentioned herein are incorporated herein by reference in their entirety. In the event of a conflict, the present specification, including definitions, shall prevail.

[0032] Unless specifically stated, the materials, methods, and examples described herein are illustrative only and not limiting. Although methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, suitable methods and materials are still described herein.

[0033] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. At the same time, the embodiments in this application and the features in the embodiments can be combined with each other without conflict.

[0034] It should be understood that any technical solution claimed for protection in the present invention does not involve the diagnosis and treatment of diseases.

[0035] In order to facilitate understanding of the embodiments of the present invention, the abbreviations and key terms that may be involved in the embodiments of the present invention are first explained or defined. For undefined abbreviations or key terms, they are commonly understood by those skilled in the art.

[0036] In addition, the test methods used in the examples are conventional methods unless otherwise specified; the materials, reagents, etc. used are all commercially available unless otherwise specified. The reagents or instruments used without indicating the manufacturer are all conventional products that can be obtained commercially. All the publications and other references mentioned in this article are incorporated herein by reference in their entirety.

[0037] The present invention is described in detail below Embodiment 1: A prediction model for the adsorption performance of micropollutants across water sources under data scarcity conditions is provided. The method for constructing the prediction model comprises the following steps: Step 1: Obtain adsorption dataset In this example, a total of 11 activated carbon samples and 7 odor-causing substances (2 earthy musty substances and 5 typical cyclic acetal substances) were collected, and adsorption isotherm experimental data in ultrapure water (UP) and local source water (NW) were obtained to generate ultrapure water adsorption data sets and source water adsorption data sets. Each data set also includes odor-causing substance concentration, pH value, and temperature. The adsorption performance is evaluated by the solid-liquid phase partition coefficient logKd, as shown in formula (1), which can better compare different equilibrium concentrations c e Under the adsorption condition, Formula (1) Five Abraham descriptors (E, S, A, B, V) were used to describe the characteristics of odor-causing substances, and the total pore volume (Vt) and BET surface area were used as descriptors of activated carbon samples. For adsorption conditions, c e is the only input feature.

[0038] Step 2: Build the base model The experimental dataset of odor-causing substances in pure water was used to develop the pedestal model. Deep neural network (DNN) was selected as the application algorithm of the transfer learning model. The optimized deep neural network consists of 5 layers, including the model input layer, three hidden layers (with 128, 64, and 32 neurons respectively) and the model output layer.

[0039] To ensure the robustness and generalizability of the model, a 5-fold cross validation was performed. To minimize data leakage, the data were grouped by different adsorbents during model development, validation, and evaluation. Model evaluation metrics included R² and root mean square error (RMSE). The trained deep neural network base model was able to extract pollutant features and make preliminary performance predictions.

[0040] Step 3: Build a transfer learning model The dataset of odorant adsorption in source water was used to develop a transfer learning model. As a method to solve the problem of data scarcity, transfer learning has become one of the main algorithms in the field of image recognition. Unlike traditional modeling methods, transfer learning does not require building a model from scratch for each new problem. It achieves the transfer and application of knowledge by pre-training the model on a large-scale dataset and then fine-tuning it on a specific problem with a small sample. This method can not only improve the accuracy of the model, but also enhance the feasibility of modeling in a small sample scenario, which is particularly applicable to the problem of data scarcity between different water sources in the field of water treatment. When fine-tuning the transfer learning model, the parameters of the base model are inherited and fixed, and only the output layer or part of the hidden layer are adjusted. A small number of samples (2-100 out of more than 600 data points) are used to fine-tune the parameters of the transfer learning model, and the remaining data points are used to evaluate the degree of understanding and generalization performance of the transfer learning model on the adsorption behavior of odorants in source water. Based on the base model, for specific types of micropollutants, the model parameters are further optimized through transfer learning to adapt to the adsorption performance prediction of specific pollutants.

[0041] Furthermore, the hidden layer of the deep neural network uses the activation function shown in the following formula (2): Formula (2) Among them, λ and α are fixed constants, usually λ = 1.0507 and α = 1.6732. The activation function in the above formula has the characteristic of self-normalization, which helps to keep the variance of the activation value stable during the training process and improve the computational efficiency. It gives the overall model excellent prediction performance when the data is extremely limited. By fine-tuning the transfer learning model with only two data points, the model can achieve an R of no less than 0.75. 2In addition, when the number of iterations of fine-tuning the transfer learning model is slightly larger, overfitting may occur, causing the existing knowledge to degenerate with overtraining. The inventor unexpectedly discovered that the use of the activation function of the above formula is conducive to significantly reducing the overfitting phenomenon, thereby improving data utilization and improving prediction performance.

[0042] Embodiment 2: Predictive performance of a model for predicting the adsorption of micropollutants across water sources under data-scarce conditions In the scenario of predicting the adsorption of odor-causing substances in source water, especially when data is limited, the transfer learning model (TL) shows superior performance. Figure 1 As shown in Figure 2, despite the small number of samples used for training, the transfer learning model still achieved excellent prediction performance on the test set of odorant adsorption in source water. Specifically, the transfer learning model was fine-tuned with only two data points and achieved an R of 0.75. 2 The transfer learning model can effectively use pre-trained knowledge to generalize to new areas. Here, the pre-trained knowledge is the base model for pure water odorant adsorption prediction, which contains extensive knowledge about the adsorption mechanism of pollutants, and the transfer learning model can adapt this knowledge to the understanding of the mechanism of odorant adsorption in source water. By fine-tuning only the output layer, the transfer learning model effectively inherits the weights and neuronal connections learned in the base model training, reducing the training requirements for source water odorant adsorption data while maintaining high prediction accuracy.

[0043] Embodiment 3: Performance comparison of transfer learning model and retrained DNN model In order to clarify the advantages and applicability of transfer learning, the study compared the performance of the transfer learning model and the retrained deep neural network (DNN) model on the water source water adsorption dataset. The results are as follows: Figure 2As shown in the figure, when the sample size is small, transfer learning outperforms the retrained DNN model across the board. During the fine-tuning process, the transfer learning model showed rapid adaptability to the new dataset. For example, after only one round of fine-tuning, the model achieved an R² value of 0.72 on the test set. After five consecutive rounds of fine-tuning, the R² value increased to 0.75, indicating that the model was able to quickly and effectively reach optimal performance. The retrained DNN model had difficulty achieving similar R² values, highlighting its reliance on larger datasets for effective training. This rapid convergence is a significant advantage of transfer learning models over traditional deep learning methods, which usually require large amounts of data and long training times to achieve similar results. For real-world applications where data collection is both time-consuming and expensive, the efficiency of transfer learning models in adapting to new data with a minimal number of fine-tuning iterations is particularly beneficial.

[0044] Embodiment 4: Resource comparison of transfer learning model and retraining DNN model Regarding the required computational resources, the retrained DNN model requires multiple iterations and a large amount of data to approach the performance level of the transfer learning model. Specifically, the performance of the DNN model does not improve significantly with the increase in the number of training iterations (20-100 times), which shows that the R² growth rate gradually decreases with the increase in the number of training times. Figure 3 On the other hand, the transfer learning model can achieve the best results with very few iterations, indicating that the transfer learning model is suitable for rapid deployment and tuning of the model. Similarly, when the number of iterations of fine-tuning the transfer learning model is slightly more (>10 times), the overfitting phenomenon also persists, indicating that the transfer learning model will also have a certain risk of overfitting, because in the process of model fine-tuning, the existing knowledge may be degraded due to overtraining.

[0045] Embodiment 5: On the basis of the foregoing embodiments, the model is further optimized to improve its prediction performance, and self-attention mechanism layers are added before and after the hidden layer of the deep neural network (DNN); the self-attention mechanism enables the model to establish connections between features at different positions, capture global dependencies, and thus provide richer feature representations; the model is allowed to dynamically focus on key parts of the input sequence instead of relying on a fixed receptive field, which increases the flexibility and adaptability of the model; the model is further helped to migrate and fuse features between different fields or tasks, allowing the model to focus on the features that are most important for prediction, improving the generalization ability of the model, and helping the model to better generalize to unseen data, thereby providing more accurate feature representations, helping to improve the prediction accuracy of the model, thereby improving the R² value, and the addition of the self-attention mechanism layer can also suppress the occurrence of overfitting, such as Figure 4 As shown in the figure, after the number of iterations of fine-tuning the transfer learning model that resulted in obvious overfitting was increased from slightly more than 10 to 50, the existing knowledge would not be sharply degraded due to fine-tuning training, thus allowing the model to maintain a high prediction accuracy. This method can more quickly and accurately predict the adsorption performance of micropollutants in water sources, providing a scientific basis for the control and treatment of pollutants.

[0046] Embodiment 6: A computer-readable storage medium is also provided, which stores a computer program executable by a processor. When the computer program is executed by the processor, it runs at least one step of the method for constructing a cross-water source micropollutant adsorption performance prediction model under the aforementioned data scarcity conditions, and can achieve the same technical effect. To avoid repetition, this embodiment will not be described in detail.

[0047] Embodiment 7: A computer device, comprising a memory, a processor, a communication interface and a communication bus; wherein the memory, the processor and the communication interface communicate with each other via the communication bus; the memory is used to store computer programs; the processor is used to execute the computer programs stored in the memory, and when the processor executes the computer program, at least one step in the method for constructing a cross-water source micropollutant adsorption performance prediction model under the aforementioned data scarcity condition is implemented, and the same technical effect can be achieved. To avoid repetition, this embodiment will not be described in detail.

[0048] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PR AM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0049] The conventional techniques in the above embodiments are prior arts known to those skilled in the art, and thus will not be described in detail here.

[0050] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0051] Although the present invention has been described in detail and some specific embodiments have been cited, it is obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the present invention.

[0052] Although the above-mentioned specific embodiments have been shown, described and pointed out the novel features applied to various embodiments, it should be understood that various omissions, substitutions and changes can be made to the form and details of the described device or method without departing from the spirit of the present disclosure. In addition, the above-mentioned various features and methods can be used independently of each other, or can be combined in various ways. All possible combinations and sub-combinations are intended to fall within the scope of the present disclosure. Many of the above-mentioned embodiments include similar components, and therefore, these similar components are interchangeable in different embodiments. Although the present invention has been disclosed in the context of certain embodiments and examples, it should be understood by those skilled in the art that the present invention can extend beyond the specifically disclosed embodiments to other alternative embodiments and / or applications and obvious modifications and equivalents thereof. Therefore, the present invention is not intended to be limited by the specific disclosure of the preferred embodiments herein.

[0053] Matters not covered in the present invention are all known technologies.

Claims

1. A method for constructing a prediction model for the adsorption performance of micropollutants across water sources, characterized in that include: Step 1, obtaining an adsorption data set: collecting at least one adsorption sample and at least one odor-causing substance, and performing adsorption in ultrapure water and source water to obtain adsorption isotherm data, so as to generate an ultrapure water adsorption data set and a source water adsorption data set; Step 2: Build a pedestal model: Develop a pedestal model using the ultrapure water adsorption data set, and use an optimized deep neural network as the application algorithm of the pedestal model. The deep neural network includes an input layer, three hidden layers, and an output layer, and a five-fold cross validation is performed; Step 3. Build a transfer learning model: Develop a transfer learning model with the source water adsorption dataset, pre-train the base model with the ultrapure water adsorption dataset, transfer the knowledge of the base model to the transfer learning model and fine-tune it, and only adjust the output layer and / or part of the hidden layer of the transfer learning model.

2. The method according to claim 1, characterized in that: In step 1, the adsorption performance is expressed as the solid-liquid partition coefficient log K d To evaluate, as shown in formula (1), Formula (1) Among them, c e Represents different equilibrium concentrations; q e It indicates the amount of chemical substance on a solid surface per unit mass or unit area when adsorption equilibrium is reached.

3. The method according to claim 1, characterized in that: In step 1, five Abraham descriptors E, S, A, B, and V were used to describe the characteristics of odor-causing substances, and total pore volume and BET surface area were used as descriptors of adsorbed samples.

4. The method according to claim 1, characterized in that: In step 2, the three hidden layers have 128, 64 and 32 neurons respectively.

5. The method according to claim 1, characterized in that: In step 2, to minimize data leakage, the data were grouped by different adsorbents during model development, validation, and evaluation.

6. The method according to any one of claims 1 to 5, characterized in that: Add self-attention mechanism layers before and after the hidden layers of the deep neural network.

7. A prediction model for the adsorption performance of micropollutants across water sources obtained based on the construction method described in any one of claims 1 to 6.

8. A method for predicting the adsorption performance of micropollutants across water sources under data scarcity conditions, characterized by: The prediction is performed based on the cross-water source micropollutant adsorption performance prediction model described in claim 7.

9. A computer device, comprising a memory, a processor, a communication interface and a communication bus; wherein: The memory, processor, and communication interface communicate with each other via the communication bus; the memory is used to store computer programs; the processor is used to execute the computer programs stored in the memory, and is characterized in that: when the processor executes the computer program, it implements the method for constructing a cross-water source micropollutant adsorption performance prediction model as described in any one of claims 1 to 6 or the method for predicting cross-water source micropollutant adsorption performance under data scarcity conditions as described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method for constructing a model for predicting the adsorption performance of micropollutants across water sources as described in any one of claims 1 to 6 or the method for predicting the adsorption performance of micropollutants across water sources under data scarcity conditions as described in claim 8.

Images