A method for accurate prediction of pollutant conversion in water treatment processes
By using a multi-sensory field attention framework for degradation path prediction, the problem of inefficient pollutant degradation path prediction in water treatment processes is solved, achieving efficient and accurate pollutant degradation path prediction, thereby improving the treatment efficiency and effluent quality of water treatment processes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-05
AI Technical Summary
In existing water treatment processes, the prediction process for pollutant degradation pathways is inefficient and the prediction results are not very accurate, which affects the design of pollutant degradation pathways and thus cannot guarantee the treatment efficiency and effluent quality of the water treatment process.
A degradation path prediction model employing a multi-receptive field attention framework acquires water treatment process parameters and pollutant molecular data. By utilizing parallel global and local attention submodules and learnable fusion operators, it captures and fuses global and local features of pollutant treatment information, thereby achieving accurate prediction of pollutant degradation paths.
It improves the prediction efficiency and accuracy of pollutant degradation pathways, ensures the treatment efficiency and effluent quality of water treatment processes, and reduces the need for human intervention and experimental design.
Smart Images
Figure CN122157870A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of pollutant conversion technology, and in particular to a method for accurately predicting pollutant conversion in water treatment processes. Background Technology
[0002] Water treatment processes are methods for removing pollutants from water or improving water quality to meet different usage or discharge standards. Pollutant degradation pathways visually demonstrate the entire process of pollutant removal in water treatment processes; therefore, designing reasonable pollutant degradation pathways can ensure the treatment efficiency and effluent quality of water treatment processes.
[0003] However, existing technologies are inefficient in predicting pollutant degradation pathways in water treatment processes and have low accuracy in predicting results, which affects the design of pollutant degradation pathways and thus cannot guarantee the treatment efficiency and effluent quality of water treatment processes. Summary of the Invention
[0004] This invention proposes a method for accurately predicting the transformation of pollutants in water treatment processes. It employs a degradation path prediction model with a multi-sensory field attention framework to predict and analyze pollutant treatment information, thereby obtaining pollutant degradation path information. The above prediction process is highly efficient and the prediction results are highly accurate, which can ensure the treatment efficiency and effluent quality of the water treatment process.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for accurately predicting the transformation of pollutants in water treatment processes, comprising: acquiring pollutant treatment information; the pollutant treatment information including water treatment process parameters and pollutant molecular data; and then determining pollutant degradation path information based on the pollutant treatment information and a degradation path prediction model; wherein, the degradation path prediction model includes a multi-receptive field attention framework; the multi-receptive field attention framework includes sequentially connected attention modules and learnable fusion operators; the attention module includes parallel global attention sub-modules and local attention sub-modules; and the pollutant degradation path information illustrates the process of pollutant transformation in the water treatment process.
[0006] In one implementation of the first aspect, pollutant degradation path information is determined based on pollutant treatment information and a degradation path prediction model. This includes: inputting pollutant treatment information into the degradation path prediction model; the degradation path prediction model includes an encoder, a multi-receptive-field attention framework, and a decoder connected in sequence. The encoder extracts initial features from the pollutant treatment information, the multi-receptive-field attention framework captures and fuses global and local features from the initial features, and the decoder then decodes and outputs the pollutant degradation path information.
[0007] In one implementation of the first aspect, a multi-receptive-field attention framework captures and fuses global and local features from the initial features, including: a global attention submodule and a local attention submodule capturing global and local features respectively; the global and local features respectively demonstrate the influence of the overall molecular structure of the pollutant and local functional groups on the degradation pathway of the pollutant. A learnable fusion operator fuses the global and local features.
[0008] In one implementation of the first aspect, the water treatment process parameters include oxidant type and concentration, pH value, catalyst type and dosage, dissolved oxygen, and / or reaction time. Pollutant molecular data includes pollutant molecular structure, atoms, bond types, and / or functional group identifiers. Pollutant degradation pathway information includes the pollutant degradation pathway, product profile, and / or key structural sites.
[0009] Secondly, this invention provides an apparatus for accurately predicting the transformation of pollutants in water treatment processes, comprising an information acquisition module and a prediction module. The information acquisition module is used to acquire pollutant treatment information, including water treatment process parameters and pollutant molecular data. The prediction module is used to determine pollutant degradation path information based on the pollutant treatment information and a degradation path prediction model; wherein the degradation path prediction model includes a multi-receptive-field attention framework; the multi-receptive-field attention framework includes sequentially connected attention modules and learnable fusion operators; the attention modules include parallel global attention sub-modules and local attention sub-modules; the pollutant degradation path information demonstrates the process of pollutant transformation in the water treatment process.
[0010] In one implementation of the second aspect, the prediction module is specifically used to input pollutant treatment information into a degradation path prediction model. The degradation path prediction model includes an encoder, a multi-receptive field attention framework, and a decoder connected in sequence. The encoder extracts the initial features of the pollutant treatment information, the multi-receptive field attention framework captures and fuses the global and local features in the initial features, and then the decoder decodes and outputs the pollutant degradation path information.
[0011] Thirdly, the present invention provides an electronic device including a processor and a memory coupled to the processor; the memory is used to store computer instructions, and when the electronic device is running, the processor executes the computer instructions stored in the memory to cause the electronic device to perform the method described in the first aspect above or any implementation thereof.
[0012] Fourthly, the present invention provides a computer-readable storage medium including computer program instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect above or any implementation thereof.
[0013] Fifthly, the present invention provides a computer program product, including computer program instructions, which, when executed on a computer, cause the computer to perform the method described in the first aspect above or any implementation thereof.
[0014] The technical effects of the second to fifth aspects and their possible implementations can be referred to the above description of the technical effects of the first aspect and its possible implementations, and will not be repeated here.
[0015] Compared with the prior art, the present invention has the following beneficial effects.
[0016] The method for predicting pollutant degradation pathways in the aforementioned water treatment process provided by this invention first acquires water treatment process parameters and pollutant molecular data as pollutant treatment information. Then, a degradation pathway prediction model with a multi-sensory-field attention framework is used to predict and analyze the pollutant treatment information to obtain pollutant degradation pathway information. During the prediction process, the global attention submodule and local attention submodule of the multi-sensory-field attention framework within the degradation pathway prediction model can extract the global and local features of the pollutant treatment information. A learnable fusion operator is then used to fuse the global and local features to output the pollutant degradation pathway information. This method not only eliminates the need for manual intervention and experimental design but also enables the prediction of pollutant degradation pathway information at the molecular level. It has the advantages of high prediction efficiency and high prediction accuracy, thereby ensuring the treatment efficiency and effluent quality of the water treatment process. Attached Figure Description
[0017] Figure 1 This is one of the schematic diagrams of a method for accurately predicting the transformation of pollutants in water treatment processes provided in the embodiments of this application; Figure 2 This is a second schematic diagram of a method for accurately predicting the transformation of pollutants in water treatment processes, provided in an embodiment of this application. Figure 3 This is one of the schematic diagrams of the degradation path prediction model provided in the embodiments of this application; Figure 4 This is the third schematic diagram of a method for accurately predicting the transformation of pollutants in water treatment processes, provided in the embodiments of this application. Figure 5 This is the second schematic diagram of the architecture of the degradation path prediction model provided in the embodiments of this application; Figure 6 This is a schematic diagram of a device for accurately predicting the transformation of pollutants in water treatment processes, provided in an embodiment of this application. Detailed Implementation
[0018] In the specification and claims of this invention, the terms "first" and "second," etc., are used to distinguish different objects, rather than to describe a specific order of objects.
[0019] In the embodiments of this application, "and / or" indicates a relationship between objects. For example, A and / or B can represent the following three situations: A exists alone, B exists alone, and A and B exist simultaneously.
[0020] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0021] The methods and apparatus provided in this application relate to water treatment processes and can be used to predict the degradation pathways of pollutants in water treatment processes.
[0022] To address the problem that existing technologies in the background are inefficient and have low accuracy in predicting pollutant degradation pathways in water treatment processes, thus affecting the design of pollutant degradation pathways and failing to guarantee the treatment efficiency and effluent quality of water treatment processes, this application provides a method for accurately predicting pollutant transformation in water treatment processes. This method employs a degradation pathway prediction model with a multi-sensory-field attention framework to predict and analyze pollutant treatment information, obtaining pollutant degradation pathway information. The prediction process is highly efficient and the prediction results are highly accurate, ensuring the treatment efficiency and effluent quality of water treatment processes.
[0023] For example, the method for accurately predicting the transformation of pollutants in water treatment processes provided in this embodiment of the invention can be executed by an electronic device with processing capabilities, such as a computer or server. Taking a computer as an example, the hardware components of the computer may include: a processor, memory, a network interface, a user interface, a communication bus, etc.
[0024] The processor controls the electronic device to perform related processing and calculation tasks, such as acquiring pollutant treatment information and determining pollutant degradation pathway information. The processor may include a central processing unit (CPU) or other processors, and may be single-core or multi-core; for example, a processor may include multiple CPUs.
[0025] Memory is used to store computer instructions and related data, such as information on pollutant treatment, degradation pathway prediction models, and pollutant degradation pathway information. Memory can be random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical storage, disk storage media, or other magnetic storage devices, or any other medium capable of storing program code or data accessible by a computer. Optionally, memory can be integrated into the processor, or it can be independent of the processor.
[0026] A network interface is used for communication between a computer and other devices or communication networks. A network interface can be a transceiver with transmit and receive capabilities. Optionally, a network interface may include standard wired interfaces or wireless interfaces (such as Wi-Fi interfaces, Bluetooth interfaces, and 5G interfaces).
[0027] The communication bus is used to enable communication between different components. For example, the processor, memory, network interface and user interface mentioned above can be interconnected through the communication bus.
[0028] The user interface may include a display screen and an input unit (such as a keyboard). Optionally, the user interface may also include a standard wired interface or a wireless interface.
[0029] Those skilled in the art will understand that the computer described above may include more or fewer components, or combine certain components, or have different component arrangements; the embodiments of this application do not limit this.
[0030] like Figure 1 As shown in the embodiments of this application, a method for accurately predicting the transformation of pollutants in water treatment processes includes S101-S102.
[0031] S101. Obtain pollutant treatment information.
[0032] In this embodiment of the application, the pollutant treatment information includes water treatment process parameters and pollutant molecular data.
[0033] Optionally, the aforementioned water treatment process parameters (also known as reaction condition information) may include oxidant type and concentration, pH value, solution temperature, light intensity, catalyst type and dosage, dissolved oxygen, reaction time and / or ionic strength, etc. The aforementioned pollutant molecular data may include pollutant molecular structure, atoms, bond types and / or functional group identifiers, etc.
[0034] Specifically, the molecular structures of the aforementioned pollutants can be represented using molecular graph structures, SMILES, and / or molecular fingerprints. The molecular graph structures represent the pollutant molecular structures using topological relationships; the SMILES represent the pollutant molecular structures using strings; and the molecular fingerprints represent the pollutant molecular structures using binary codes.
[0035] S102. Based on pollutant treatment information and degradation pathway prediction models, determine pollutant degradation pathway information.
[0036] The aforementioned degradation pathway prediction model includes a multi-receptive-field attention framework. This framework comprises sequentially connected attention modules and a learnable fusion operator. Each attention module includes parallel global attention sub-modules and local attention sub-modules. The pollutant degradation pathway information exemplifies the pollutant transformation process in water treatment. This information may include the pollutant's degradation pathway, product profile, and / or key structural sites.
[0037] In one implementation, combined with Figure 1 ,like Figure 2 As shown, S102 includes S1021-S1022.
[0038] S1021. Input pollutant treatment information into the degradation pathway prediction model.
[0039] refer to Figure 3 The degradation path prediction model mentioned above includes an encoder, a multi-receptive field attention framework, and a decoder connected in sequence.
[0040] In one application scenario, the above-mentioned pollutant treatment information needs to undergo the following preprocessing operations before being input into the degradation path prediction model.
[0041] Operation 1: Convert the above pollutant treatment information into a unified input sequence.
[0042] In some implementations, the input sequence consists of multiple tokens, where each token refers to the smallest processing unit of the degradation path prediction model.
[0043] The tokens mentioned above can be categorized into structural tokens and conditional tokens. Structural tokens are used to encode local structural units of molecules (e.g., atoms, bonds, segments, substructures, functional groups, or combinations thereof); conditional tokens are used to encode reaction conditions (e.g., through discretization embedding, numerical feature projection, interval encoding, or combinations thereof).
[0044] Operation 2: Perform position encoding and mask construction on the input sequence to obtain the preprocessed input sequence.
[0045] The aforementioned positional encoding is used to represent the relative and absolute positional information of tokens in the sequence; the aforementioned mask includes at least a padding mask for identifying invalid padding positions. In some embodiments, it may also include a type mask or other task-related masks for distinguishing between structure tokens and condition tokens. Since positional encoding and mask construction are common techniques in this technical field, they will not be described in detail here.
[0046] S1022. The encoder extracts the initial features of pollutant treatment information. The multi-receptive field attention framework captures the global and local features in the initial features and fuses them. The decoder then outputs the pollutant degradation path information after decoding.
[0047] In one embodiment of S1022 above, the process of determining pollutant degradation pathway information includes the following steps 1-4.
[0048] Step 1: Map the above input sequence to an initial embedding representation and input it into the encoder to obtain the initial feature X output by the encoder.
[0049] For example, the encoder described above can be an attention-based encoder, and the initial feature X output by the encoder can be a sequence feature matrix of length n. The encoder can also be other encoders that implement step 1 above; this embodiment of the application does not limit the specific implementation.
[0050] Step 2: The initial feature X enters both the global attention submodule and the local attention submodule. The global attention submodule captures the global features in the initial feature X, and the local attention submodule captures the local features in the initial feature X.
[0051] For the global attention submodule mentioned above, the global features reflect the long-distance dependence of the entire sequence. When analyzing the initial feature X, the global features show the influence of the overall molecular structure of the pollutant on the degradation pathway of the pollutant.
[0052] The process by which the global attention submodule captures the global features in the initial feature X includes the following steps 2.1-2.4.
[0053] Step 2.1: Map the initial feature X to the query, key, and value spaces. The mapping formula is shown below.
[0054] in, X Let X be the initial feature. Q For the query vector sequence, K For the key vector sequence, V For the value vector sequence,W Q , W K , W V For learnable linear projection parameters, b Q , b K 、b V This is the bias term corresponding to the linear projection.
[0055] Step 2.2: Calculate the scaled dot product similarity score for each attention head m. The formula for calculating the scaled dot product similarity score is shown below.
[0056] in, S (m) Score the attention of the m-th attention head. d h Here, m is the scaling parameter and the attention head index.
[0057] Step 2.2: Apply a padding mask to block invalid locations.
[0058] Score the attention corresponding to the filling position. Set it to a minimum value so that its weights approach 0 after softmax normalization, thereby reducing the impact of noise.
[0059] Step 2.3: Softmax normalize the attention scores to obtain the attention weights. The normalization formula for the attention weights is shown below.
[0060] in, A (m) Let be the attention weight of the m-th attention head.
[0061] Step 2.4, for V Weighted convergence yields the output of each attention head, which is then concatenated and projected to obtain the global features. .
[0062] The above global features The calculation formula is shown below.
[0063] Where h is the number of attention heads, W O To output the projected weights, b O This is the output projection bias.
[0064] For the aforementioned local attention submodule, the local features focus on the details of the local window neighborhood. When analyzing the initial feature X, the local features demonstrate the influence of the local functional groups of the pollutant on the degradation pathway of the pollutant.
[0065] The process by which the local attention submodule captures local features in the initial feature X includes the following steps 2.5-2.7.
[0066] Step 2.5, based on window radius r Construct a local window visibility mask , to query position i Only able to focus on satisfaction | i - j |≤ r key position j .
[0067] The above local window visibility mask It satisfies the following formula.
[0068] Step 2.6: Overlay mask constraints in local attention calculation.
[0069] In some implementations, the similarity score for locations outside the window is set to a minimum value to force attention to normalize only within the local neighborhood; at the same time, a padding mask is used to prevent invalid tokens from being included in the local window calculation.
[0070] Step 2.7: Use the formulas in Steps 2.1 to 2.4 above to implement Q / K / V mapping, similarity calculation and normalization of local attention. Since the normalization range is limited by the local window, local features are calculated.
[0071] In some implementations, the local window radius r These are preset hyperparameters that can be adaptively set based on sequence length, molecular size, or number of tokens. r .
[0072] Step 3: Learn the fusion operator to fuse global and local features.
[0073] Understandably, the aforementioned learnable fusion operator includes a gating network. Instead of employing a static weight allocation strategy, this learnable fusion operator dynamically learns how to fuse features from global and local attention branches through a lightweight neural network. Since the aforementioned learnable fusion operator is prior art, this application will not provide further explanation of it in its embodiments.
[0074] In one implementation, step 3 above includes steps 3.1 to 3.3.
[0075] Step 3.1: Construct the residual enhancement representation.
[0076] The aforementioned global and local features are respectively residually connected to the initial feature X to obtain two enhanced representations, which preserve basic semantic information and stabilize gradient propagation.
[0077] Step 3.2: Generate fusion features.
[0078] The fusion feature is obtained by weighting and summing the two enhanced representations according to the fusion weights. O Fusion features O The calculation formula is shown below.
[0079] in, For global weights, For local weights, A This is an enhanced representation of global branch residuals. B This represents the enhancement of local branch residuals.
[0080] Step 4: After fusing the features into the decoder, the decoder outputs pollutant degradation path information.
[0081] Optionally, the pollutant degradation pathway information may include, but is not limited to: (a) the reaction pathway of the pollutant under a given operating condition (e.g., step-by-step reaction sequence, key intermediate sequence); (b) the set of possible products and their probability distribution; and (c) the contribution information of key structural sites associated with the reaction conditions (e.g., interpretive output that can be derived from attention weights).
[0082] As can be seen from the above, the core of the aforementioned method lies in the multi-receptive field attention framework. This framework, through collaborative global attention submodules, local attention submodules, and learnable fusion operators, enables deep analysis and feature extraction of the input sequence (which encodes characteristic information such as pollutant molecular structure and reaction conditions). The global attention submodule models long-distance dependencies in the sequence, capturing global association patterns across subsequences. The local attention submodule, by constraining the attention receptive field within a preset window radius, focuses on detailed patterns between adjacent positions, enhancing the ability to characterize local structures (such as key functional groups). Through this multi-scale feature capture design, the model can simultaneously encode fine-grained information about pollutant molecular structure and large-scale contextual dependencies, effectively alleviating the information compression and loss problems faced by traditional models with limited representation capacity, and improving the overall representation ability of the input sequence.
[0083] Specifically, after initial feature extraction, the model performs local and global attention computations in parallel. For the local attention submodule, a window visibility mask is defined to force each query position to focus only on key positions within its neighborhood radius, thus confining the attention mechanism to a local window. This is further combined with a padding mask to filter out invalid positions, ultimately generating an attention map and feature representation focused on local details. For the global attention submodule, the global correlation between the query position and all key positions in the sequence is calculated. Similarly, after applying a padding mask, an attention map and global feature representation reflecting the information interaction across the entire sequence are generated.
[0084] Optionally, combined Figure 2 ,like Figure 4 As shown, before S101, the method further includes S103.
[0085] S103. Construct and train a degradation pathway prediction model.
[0086] The following section uses the ultraviolet-induced degradation process of pollutants as an application scenario to train the degradation pathway prediction model.
[0087] (1) The UVDB dataset was used as the source of training and testing data.
[0088] The UVDB dataset includes multiple sample datasets, each containing: a) molecular structure information of the substrate pollutant; b) UV reaction conditions (e.g., illumination conditions, oxidation system parameters, etc.); and c) corresponding reaction pathway sequence labels, used for supervised training and evaluation.
[0089] (2) Construct the base model, referring to Figure 5 The aforementioned base model is a base model that embeds the aforementioned multi-receptive field attention framework between the encoder and decoder provided by Google's BERT model.
[0090] (3) The prediction task is defined as: given the substrate molecular structure and reaction conditions, predict the possible reaction pathways of the pollutant under the target ultraviolet conditions and the corresponding product spectrum.
[0091] During the training of the basis model on the UVDB dataset, the performance of the basis model on the reaction path prediction task was evaluated using metrics such as accuracy, BLEU, and ROUGE, so as to obtain the base model after training and use the trained base model as the degradation path prediction model.
[0092] Experimental results show that on the UVDB dataset, the model achieves an ACC of 87.18%, a BLEU of 0.9459, and a ROUGE of 0.9544. These results demonstrate that the degradation pathway prediction model provided in this application exhibits high prediction accuracy and good generalization ability under different data distribution scenarios, and can meet the needs of rapid prediction of pollutant degradation pathways in AOP processes.
[0093] In summary, the method for predicting pollutant degradation pathways in the water treatment process provided in this application, employing the aforementioned multi-receptive-field attention framework, can efficiently and rapidly simulate and predict the specific reaction pathways and possible product spectra of pollutants in advanced oxidation processes under different operating conditions (such as oxidant type, concentration, pH value, light intensity, etc.), providing directly applicable prediction rules for process design and control. Secondly, thanks to the multi-receptive-field design, the model can simultaneously consider the influence of local reactive sites (fine-grained information) and the overall molecular structure on the reaction pathway during prediction, thereby achieving higher prediction accuracy, which has been verified by the high accuracy on a self-built dataset. Furthermore, by analyzing the attention map generated by the model, specific groups in the substrate molecule that play a key role in the reaction can be identified, and these groups can be associated with specific reaction conditions, achieving a certain degree of interpretable, expert-like chemical reasoning. Finally, this framework is lightweight and modular, possessing good generalization ability, and can support high-throughput pollutant degradation pathway screening, fundamentally solving the problems of high cost, long cycle, and high technical threshold of traditional methods, providing strong technical support for improving wastewater treatment efficiency and promoting environmental sustainability. .
[0094] Accordingly, embodiments of this application provide a device for accurately predicting the transformation of pollutants in water treatment processes, such as... Figure 6 As shown, it includes an information acquisition module 501 and a prediction module 502.
[0095] The information acquisition module 501 is used to acquire pollutant treatment information, including water treatment process parameters and pollutant molecular data. For example, the information acquisition module 501 is used to implement step S101 of the above method.
[0096] The prediction module 502 is used to determine pollutant degradation path information based on pollutant treatment information and a degradation path prediction model. The degradation path prediction model includes a multi-receptive-field attention framework, which comprises sequentially connected attention modules and a learnable fusion operator. Each attention module includes parallel global attention sub-modules and local attention sub-modules. The pollutant degradation path information displays the pollutant transformation process in the water treatment process. For example, the prediction module 502 is used to implement step S102 of the above method.
[0097] Optionally, the prediction module 502 is specifically used to: input pollutant treatment information into a degradation path prediction model; the degradation path prediction model includes an encoder, a multi-receptive-field attention framework, and a decoder connected in sequence; the encoder extracts initial features from the pollutant treatment information, the multi-receptive-field attention framework captures and fuses global and local features from the initial features, and then the decoder decodes and outputs pollutant degradation path information. For example, the prediction module 502 is specifically used to implement S1021-S1022 of the above method.
[0098] Each module of the pollutant degradation pathway prediction device in the above water treatment process can also be used to perform other steps in the above method embodiments. All relevant contents involved in the above method embodiments can be referred to the functional description of the corresponding functional module, and will not be repeated here.
[0099] This application also provides an electronic device, including: a processor and a memory coupled to the processor; the memory is used to store computer instructions, and when the electronic device is running, the processor executes the computer instructions stored in the memory to cause the electronic device to perform the methods in the above embodiments. The processor can implement the information acquisition module 501 and the prediction module 502 described above; the memory can also be used to store pollutant treatment information, degradation path prediction models, and pollutant degradation path information, etc.
[0100] This application also provides a computer-readable storage medium including a computer program that, when run on a computer, performs the methods described in the above embodiments.
[0101] This application also provides a computer program product, which includes computer program instructions that, when run on a computer, execute the methods described in the above embodiments.
[0102] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for accurately predicting the transformation of pollutants in water treatment processes, characterized in that, include: Obtain information on pollutant treatment; The pollutant treatment information includes water treatment process parameters and pollutant molecular data; Based on the pollutant treatment information and degradation pathway prediction model, pollutant degradation pathway information is determined; wherein, the degradation pathway prediction model includes a multi-receptive field attention framework; the multi-receptive field attention framework includes sequentially connected attention modules and learnable fusion operators; the attention module includes parallel global attention sub-modules and local attention sub-modules; The pollutant degradation pathway information demonstrates the process of pollutant transformation in water treatment technology.
2. The method as described in claim 1, characterized in that, The determination of pollutant degradation pathway information based on the pollutant treatment information and degradation pathway prediction model includes: The pollutant treatment information is input into the degradation path prediction model; the degradation path prediction model includes an encoder, the multi-receptive field attention framework, and a decoder connected in sequence. The encoder extracts initial features of the pollutant treatment information, the multi-receptive-field attention framework captures global and local features from the initial features and fuses them, and then the decoder outputs the pollutant degradation path information after decoding.
3. The method as described in claim 2, characterized in that, The multi-receptive-field attention framework captures and fuses global and local features from the initial features, including: The global attention submodule and the local attention submodule capture the global features and the local features, respectively; the global features and the local features respectively demonstrate the influence of the overall molecular structure of the pollutant and the local functional groups on the degradation pathway of the pollutant; The learnable fusion operator fuses the global features and the local features.
4. The method as described in claim 1 or 3, characterized in that, The water treatment process parameters include oxidant type and concentration, pH value, catalyst type and dosage, dissolved oxygen and / or reaction time; The pollutant molecular data includes the pollutant molecular structure, atoms, bond types, and / or functional group identifiers; The pollutant degradation pathway information includes the pollutant degradation pathway, product profile, and / or key structural sites.
5. A device for accurately predicting the transformation of pollutants in water treatment processes, characterized in that, It includes an information acquisition module and a prediction module; The information acquisition module is used to acquire pollutant treatment information; the pollutant treatment information includes water treatment process parameters and pollutant molecular data. The prediction module is used to determine pollutant degradation path information based on the pollutant treatment information and the degradation path prediction model; wherein, the degradation path prediction model includes a multi-receptive field attention framework; the multi-receptive field attention framework includes sequentially connected attention modules and learnable fusion operators; the attention module includes parallel global attention sub-modules and local attention sub-modules; the pollutant degradation path information shows the process of pollutant transformation in the water treatment process.
6. The apparatus as claimed in claim 5, characterized in that, The prediction module is specifically used to input the pollutant treatment information into the degradation path prediction model; the degradation path prediction model includes an encoder, a multi-receptive field attention framework, and a decoder connected in sequence; the encoder extracts the initial features of the pollutant treatment information, the multi-receptive field attention framework captures and fuses the global and local features in the initial features, and then the decoder decodes and outputs the pollutant degradation path information.
7. An electronic device, characterized in that, The device includes a processor and a memory coupled to the processor; the memory is used to store computer instructions, which, when the electronic device is running, are executed by the processor to cause the electronic device to perform the method as described in any one of claims 1 to 4.
8. A computer-readable storage medium, characterized in that, It includes computer program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1 to 4.
9. A computer program product, characterized in that, It includes computer program instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 4.