Iron-aluminum coagulant applicability evaluation method based on water quality zoning

Abstract

The application provides a kind of iron and aluminum coagulant applicability evaluation method based on water quality zoning, belongs to the field of iron and aluminum coagulant evaluation.The method comprises the following steps: collecting multi-dimensional data of water body, and generating standardized optical data set through purification, feature extraction and enhancement processing;using dynamic density clustering algorithm, delimiting water quality zoning boundary, and associating historical decision knowledge for each water quality zoning;constructing a corresponding mixed intelligent evaluation model for each dynamic water quality zoning, and establishing a multi-objective optimization decision function to obtain the optimal coagulant decision scheme;constructing a lightweight digital twin model corresponding to the target water plant area, and simulating and deducing the optimal coagulant decision scheme.The application realizes the precise matching of iron and aluminum coagulant adaptability, improves the pertinence of coagulant selection and dosing strategy, reduces the blindness of model decision, ensures the stability of coagulation treatment effect, and improves the adaptability of mixed intelligent evaluation model to various special water quality scenarios.

Description

Technical Field

[0001] This invention relates to the field of iron-aluminum coagulant evaluation technology, and more specifically to a method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning. Background Technology

[0002] Iron-aluminum coagulants are widely used in coagulation processes for municipal water supply and industrial wastewater treatment due to their stable coagulation effect and controllable cost. Their applicability evaluation and dosing strategy optimization are key to improving water treatment efficiency and reducing operation and maintenance costs. A water quality zoning-based applicability evaluation method for iron-aluminum coagulants, by dividing areas with similar water quality characteristics and matching them with specific coagulant dosing schemes, represents a core approach to breaking through the traditional uniform dosing model.

[0003] In existing technologies, the applicability evaluation method for iron-aluminum coagulants based on water quality zoning still has the following drawbacks: First, it has poor adaptability to dynamic fluctuations in water quality. Fixed zoning standards cannot respond to seasonal changes and sudden pollution, leading to a lag in the selection and dosing strategies of iron-aluminum coagulants, resulting in unstable treatment effects and waste of reagents. Second, the evaluation model performance is unbalanced. Pure mechanistic models are difficult to fit nonlinear relationships under complex water quality disturbances, while pure AI models lack theoretical support and their decisions are uninterpretable, failing to balance the scientific nature and accuracy of the evaluation. Third, the model has insufficient generalization ability and data privacy is difficult to guarantee. The limited amount of data from a single water plant results in a narrow range of applicable scenarios for the model. Data sharing across water plants is prone to leakage of process and data privacy, making it impossible to achieve collaborative reuse of technical knowledge from multiple parties, thus limiting the large-scale promotion of this type of evaluation method. Summary of the Invention

[0004] The purpose of this invention is to provide a method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning, in order to solve the technical problems in the prior art, such as poor adaptability of iron-aluminum coagulants when water quality fluctuates dynamically, lagging zoning decision-making, difficulty in coagulant evaluation models to balance mechanism interpretability and fitting accuracy in complex scenarios, and insufficient generalization ability of models due to the scarcity of data from a single water plant.

[0005] To achieve the above objectives, this invention provides a method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning. The method includes: collecting multi-dimensional water data from multimodal sensors deployed in the target water plant area; generating a standardized optical dataset through purification, feature extraction, and enhancement processing; transforming the standardized optical dataset from high-dimensional data to low-dimensional space using a self-organizing mapping network; delineating water quality zoning boundaries using a dynamic density clustering algorithm; associating historical decision knowledge with each water quality zoning zone to generate water quality zoning labels and zoning maps; constructing a corresponding hybrid intelligent evaluation model for each dynamic water quality zoning zone based on the standardized optical dataset and water quality zoning labels; establishing a multi-objective optimization decision function; solving for the optimal coagulant decision scheme; and using digital twin technology to construct a lightweight digital twin model corresponding to the target water plant area, simulating the optimal coagulant decision scheme, and comparing the simulation results with preset water quality compliance requirements for feedforward optimization and strategy calibration.

[0006] Optionally, the process of collecting multi-dimensional water body data and generating a standardized optical dataset through purification, feature extraction, and enhancement includes: employing an adaptive noise reduction algorithm to suppress noise in the multi-dimensional water body data, eliminating noise generated by equipment operation and environmental interference, and retaining the detailed features of the effective spectrum; performing internal filtration effect correction on the noise-suppressed multi-dimensional water body data, constructing a quantitative correlation between water concentration and fluorescence intensity attenuation, and performing point-by-point correction on the spectral signal; subtracting Raman scattering from the point-by-point corrected spectral signal, using baseline correction to remove Raman scattering interference, highlighting the inherent optical features in the water body, and obtaining purified water body spectral data; performing time-frequency analysis on the purified water body spectral data using wavelet transform, and using principal component analysis to reduce the dimensionality of the high-dimensional wavelet feature data, extracting target optical feature indicators, and generating a standardized optical dataset.

[0007] Optionally, the step of employing a dynamic density clustering algorithm to delineate water quality zoning boundaries and associating historical decision knowledge with each water quality zoning zone includes: based on the low-dimensional topological grid clustering results generated by the self-organizing map network, using a dynamic density clustering algorithm, setting a density threshold, and identifying high-density core areas and low-density boundary areas of water quality feature data points in the low-dimensional topological grid; when the influent water quality changes, the distribution density of water quality feature data points changes accordingly, and the dynamic density clustering algorithm automatically identifies the changes and dynamically adjusts the range and number of zoning boundaries; individually marking the identified water quality anomalies to ensure that the zoning results truly reflect the water quality status; assigning a corresponding zoning identifier to each identified stable water quality zoning zone and constructing a dynamically updated zoning-decision knowledge base, associating each zoning identifier with the corresponding historical decision knowledge; establishing a fast query index for zoning and decision knowledge, and quickly calling the corresponding historical optimal decision knowledge in the knowledge base through the zoning identifier.

[0008] Optionally, the step of constructing a corresponding hybrid intelligent evaluation model for each dynamic water quality zone based on standardized optical datasets and water quality zone labels includes: constructing the mechanism kernel of the hybrid intelligent evaluation model according to the preset coagulation reaction mechanism; constructing a deep neural network as the AI ​​driving module of the hybrid intelligent evaluation model; training the hybrid intelligent evaluation model with historical verification data as supervision signals; and using an adaptive moment estimation algorithm to adjust the weights and bias parameters of the deep neural network and introducing a regularization mechanism to optimize and correct the calculation results of the hybrid intelligent evaluation model through multiple rounds of iterative training.

[0009] Optionally, the step of using digital twin technology to construct a lightweight digital twin model corresponding to the target water plant area and simulating the optimal coagulant decision scheme includes: the lightweight digital twin model covering the process units of the entire coagulation treatment process, ensuring consistency between the lightweight digital twin model and the physical process units; loading the optimal coagulant decision scheme into the lightweight digital twin model and performing millisecond-level full-process simulation to recreate the dynamic process of coagulation treatment; predicting the water quality index values ​​of each target node during the simulation and comparing them with the preset water quality compliance requirements to identify process adaptation issues in the optimal coagulant decision scheme.

[0010] Optionally, the iron-aluminum coagulant suitability evaluation method further includes: using federated learning technology, treating all target water plant areas applying the iron-aluminum coagulant suitability evaluation method as independent nodes, deploying a corresponding hybrid intelligent evaluation model on each independent node, performing local model incremental training, calculating the gradient information of model parameters, and realizing the collaborative updating of the hybrid intelligent evaluation model.

[0011] Optionally, the step of performing local incremental model training and calculating the gradient information of model parameters includes: conducting local incremental model training based on the new data generated by each node running locally; during the training process, retaining the core mechanism of the hybrid intelligent evaluation model and incrementally optimizing the parameters of the AI-driven module; setting a corresponding training log for each node to record parameter adjustments and model performance changes during the training process; after the incremental training is completed, calculating the update amount of model parameters as gradient information and encrypting the gradient information using a homomorphic encryption algorithm; constructing a preset secure data transmission channel, using an encrypted transmission protocol, restricting transmission permissions, and transmitting the encrypted gradient information to the central aggregation server.

[0012] Optionally, the method for evaluating the suitability of iron-aluminum coagulants further includes: establishing a communication connection with the controller of the target water plant area through an industrial Internet of Things interface, connecting the optimal coagulant decision scheme with the actual process execution, and deploying an online water quality monitoring instrument at the end of the water coagulation and sedimentation process to monitor changes in effluent water quality and conduct dynamic water quality early warning and closed-loop feedback.

[0013] Optionally, the deployment of online water quality monitoring instruments monitors changes in effluent water quality and performs dynamic water quality early warning and closed-loop feedback, including: constructing an adaptive dynamic early warning threshold system based on the fluctuation pattern of influent water quality, historical treatment data, and preset water quality compliance requirements; when the deviation between the monitored effluent water quality indicators and the predicted values ​​of the lightweight digital twin model continuously exceeds the dynamic early warning threshold, or when the effluent indicators reach the water quality compliance requirement limit, a graded early warning is triggered, and early warning information is pushed to the operation and maintenance personnel; after the dynamic water quality early warning is triggered, it automatically determines whether the water quality or process unit has changed, and feeds back the monitoring data and early warning information to the beginning of the evaluation process for a new round of applicability evaluation.

[0014] Optionally, the method for evaluating the suitability of iron-aluminum coagulants also includes: establishing a routine operation and maintenance mechanism, calibrating the accuracy of multimodal sensors and monitoring instruments according to a preset cycle, conducting a full-process data review and analysis of the iron-aluminum coagulant suitability evaluation system, and combining the review and analysis results with the federated learning collaborative update results to perform iterative optimization of all modules of the system.

[0015] Through the above technical solutions, this invention solves the problem that traditional fixed zoning cannot adapt to dynamic changes in water quality by constructing a dynamic water quality zoning map and a zoning-decision knowledge base. This achieves precise matching of the compatibility of iron-aluminum coagulants, avoids unstable treatment effects caused by lagging zoning decisions, improves the targeting of coagulant selection and dosing strategies, and reduces reagent waste and the risk of effluent compliantness. By constructing a hybrid intelligent evaluation model and a digital twin system, it addresses the industry pain points of pure mechanistic models being unable to cope with complex water quality interference and pure AI model decisions being unexplainable. This improves the scientificity and accuracy of iron-aluminum coagulant dosing strategies while reducing the blindness of model decisions and ensuring the stability of coagulation treatment effects. Through federated learning technology and homomorphic encryption to ensure gradient transmission security, it solves the problems of limited data volume in a single water plant and narrow model adaptation scenarios. Without disclosing the process and data privacy of each water plant, it integrates water quality treatment knowledge from the entire network, improves the adaptability of the hybrid intelligent evaluation model to various special water quality scenarios, and expands the scope of technology application.

[0016] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0018] Figure 1 This is a flowchart illustrating the method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to the present invention. Figure 1 ;

[0019] Figure 2 This is a schematic diagram of the process for generating a standardized optical dataset in this invention;

[0020] Figure 3 This is a schematic diagram of the process for defining water quality zoning boundaries in this invention;

[0021] Figure 4 This is a schematic diagram of the process for constructing the corresponding hybrid intelligent evaluation model in this invention;

[0022] Figure 5 This is a schematic diagram of the process for constructing a lightweight digital twin model in this invention;

[0023] Figure 6 This is a schematic diagram of the process of incremental training of the local model in this invention;

[0024] Figure 7 This is a schematic diagram of the process for dynamic water quality early warning and closed-loop feedback in this invention;

[0025] Figure 8 This is a flowchart illustrating the method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to the present invention. Figure 2 . Detailed Implementation

[0026] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0027] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0028] Please refer to Figure 1 and Figure 8 This invention provides a method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning. The method may include:

[0029] Step S100: Collect multi-dimensional data of the water body based on the multi-modal sensors deployed in the target water plant area, and generate a standardized optical dataset through purification, feature extraction and enhancement processing.

[0030] In this embodiment of the invention, based on the characteristics of the target water plant's process flow and the need for full coverage of water quality monitoring, integrated multi-parameter monitoring units can be scientifically deployed at key process nodes such as the inlet, pretreatment front end, and sedimentation tank inlet and outlet to construct a multimodal sensor network with full coverage and high-frequency synchronization. Each monitoring unit synchronously integrates an ultraviolet-visible full-spectrum sensor, a three-dimensional fluorescence spectral sensor, and conventional physicochemical parameter sensors. The ultraviolet-visible full-spectrum sensor can capture spectral signals in the wavelength range of 200-800nm, the three-dimensional fluorescence spectral sensor focuses on the fluorescence response characteristics of dissolved organic matter in the water, and the conventional physicochemical sensors synchronously collect basic indicators such as pH, temperature, turbidity, and conductivity to comprehensively characterize the overall properties of the water body. After deployment, all sensors are uniformly debugged and calibrated to calibrate the signal synchronization of each sensor and set the same high-frequency acquisition frequency to avoid data time axis misalignment caused by signal delay.

[0031] Please refer to Figure 2 In this embodiment of the invention, multi-dimensional water body data is collected, and a standardized optical dataset is generated through purification, feature extraction, and enhancement processing, which may include:

[0032] Step S110: Adaptive noise reduction algorithm is used to suppress noise in multi-dimensional water data, eliminate noise generated by equipment operation and environmental interference, and retain the detailed features of the effective spectrum.

[0033] In a preferred embodiment of the present invention, the acquired raw spectral data is susceptible to multiple interferences, including equipment noise, water turbidity scattering, Raman scattering, and internal filtration effects, making it unsuitable for direct feature extraction. Therefore, a multi-step raw purification process is required to remove irrelevant interference signals to the greatest extent possible, preserving the pure optical characteristics of dissolved organic matter in the water. An adaptive denoising algorithm is employed to suppress noise in the raw spectral data. By analyzing the amplitude fluctuation patterns of the spectral signal, the difference between random noise and the effective signal is accurately identified. The adaptive denoising algorithm is then used to specifically remove noise generated by equipment operation and environmental interference, while preserving the detailed features of the effective spectrum and avoiding feature loss due to excessive denoising.

[0034] Step S120: Perform internal filtration effect correction on the multi-dimensional water data after noise suppression, construct a quantitative correlation between water concentration and fluorescence intensity attenuation, and perform point-by-point correction on the spectral signal.

[0035] In a preferred embodiment of the present invention, in order to address the problem of fluorescence intensity attenuation caused by the scattering and absorption of light by organic matter or particulate matter in high-concentration water samples, a quantitative correlation between water concentration and fluorescence intensity attenuation is constructed. Combined with turbidity data from conventional physicochemical parameters, the spectral signal is corrected point by point to restore the true fluorescence signal intensity of the water.

[0036] Step S130: Raman scattering is subtracted from the point-by-point calibrated spectral signal, and the Raman scattering interference is removed by baseline correction method to highlight the inherent optical characteristics of the water body and obtain the purified water spectral data.

[0037] In a preferred embodiment of the present invention, by comparing the Raman scattering spectral characteristics of pure water, interference peaks generated by Raman scattering in the original spectrum are accurately identified, and the baseline correction method is used to remove such interference, further purifying the spectral data and highlighting the inherent optical characteristics of different types of organic matter in the water.

[0038] Step S140: Wavelet transform is used to perform time-frequency analysis on the spectral data of the purified water body, and principal component analysis is used to reduce the dimensionality of the high-dimensional wavelet feature data, extract the target optical features (e.g., fluorescence intensity ratio, fluorescence index, humification index, biogenic index and other key optical feature indicators) and generate a standardized optical dataset.

[0039] In a preferred embodiment of the present invention, wavelet transform is used to perform time-frequency analysis on the purified spectrum. Wavelet basis functions suitable for the spectral signal are selected, and the complex spectral signal is decomposed into components of different frequencies through multi-scale decomposition. The low-frequency components correspond to the characteristic spectral peaks of organic matter, and the high-frequency components correspond to residual weak interference. By retaining the low-frequency components and filtering the high-frequency residual interference, the ability to characterize the composition and structure of different types of organic matter in the water is enhanced, and the spectral response differences of different organic matter such as humic acid and protein are accurately captured. Principal component analysis is used to reduce the dimensionality of the high-dimensional wavelet feature data. Through linear transformation, multiple highly correlated spectral feature indicators are integrated into a few independent principal components. The principal components with the required cumulative contribution rate are selected to retain the core information of the original spectrum to the maximum extent.

[0040] For example, an integrated monitoring buoy was deployed at a monitoring section of the Yellow River in Ningxia, constructing a multimodal sensor network to simultaneously collect ultraviolet-visible full-spectrum data, three-dimensional fluorescence spectra, and conventional physicochemical parameters. During monitoring, the sensors captured significant humic acid-like fluorescence signals and protein-like fluorescence signals. The raw spectral data contained slight equipment noise and turbidity interference. After filtering noise using an adaptive denoising algorithm, internal filtration effect correction and Raman scattering subtraction were performed to purify the spectral signals. Then, wavelet transform was used to extract multi-scale features, combined with principal component analysis for dimensionality reduction, ultimately refining core optical characteristic indicators such as fluorescence index and humification index. The processed data showed that the fluorescence index of this water body conformed to the characteristics of terrestrial organic matter input, clarifying the main source of organic matter in the water.

[0041] Step S200: Using a self-organizing mapping network, the standardized optical dataset is transformed from high-dimensional data to low-dimensional space. A dynamic density clustering algorithm is used to delineate the water quality zoning boundaries and associate historical decision knowledge with each water quality zoning to generate water quality zoning labels and zoning maps.

[0042] In this embodiment of the invention, the self-organizing map network simulates the topological structure of brain nerve cells, consisting of an input layer and a competition layer. The input layer is connected to a standardized optical dataset, and the competition layer consists of a low-dimensional topological grid composed of multiple neurons. Each neuron corresponds to a set of weight vectors, consistent with the dimension of the input feature vector. During training, the network conducts unsupervised competitive learning. For each set of input water quality feature data, the similarity between it and the weight vectors of all neurons in the competition layer is calculated. The neuron with the highest similarity is selected as the winning neuron, and the weight vectors of the winning neuron and its neighboring neurons are updated, so that the weight vectors of the neighboring neurons gradually converge towards the input data. After multiple rounds of iterative training, the high-dimensional nonlinearly distributed water quality feature data is accurately projected onto the low-dimensional topological grid. Data with similar water quality features are mapped to adjacent neurons, and data with different water quality features are assigned to different regions of the grid, clearly presenting the natural clustering trend of water quality features.

[0043] Please refer to Figure 3 In this embodiment of the invention, a dynamic density clustering algorithm is used to delineate water quality zoning boundaries and associate historical decision knowledge with each water quality zoning zone, which may include:

[0044] Step S210: Based on the low-dimensional topological grid clustering results generated by the self-organizing map network, a dynamic density clustering algorithm is used to set a density threshold and identify the high-density core region and low-density boundary region of water quality characteristic data points in the low-dimensional topological grid.

[0045] In a preferred embodiment of the present invention, the dynamic density clustering algorithm does not require a pre-set number of clusters. It relies on the distribution density of data points to carry out clustering. By setting a density threshold, it automatically identifies the high-density core region and low-density boundary region of water quality characteristic data points in the low-dimensional grid. The high-density core region is a stable water quality partition with similar water quality characteristics and small fluctuations, while the low-density region is the transition boundary of different partitions or water quality anomaly points.

[0046] Step S220: When the influent water quality changes (e.g., seasonal changes, sudden pollution, or gradual changes in water quality), the distribution density of water quality feature data points changes accordingly. The dynamic density clustering algorithm automatically identifies the changes and dynamically adjusts the range and number of partition boundaries.

[0047] Step S230: Individually mark the identified water quality anomalies to ensure that the zoning results truly reflect the water quality status.

[0048] In a preferred embodiment of the present invention, the identifier adopts a combination of letters and numbers, which is convenient for distinguishing different water quality zones and can also be associated with the core characteristics of the zone (e.g., water quality type, main pollutants), thereby improving the zone's recognizability.

[0049] Step S240: Assign a corresponding partition identifier to each identified stable water quality partition, and construct a dynamically updated partition-decision knowledge base, associating each partition identifier with the corresponding historical decision knowledge.

[0050] In a preferred embodiment of the present invention, the partition-decision knowledge base adopts a structured data storage mode, which accurately associates each partition identifier with the corresponding historical monitoring data and historical coagulation treatment data. The integrated core decision information includes the best-performing coagulant type in the partition (e.g., single coagulant or compound coagulant), the optimal range of reagent addition parameters, treatment effect data (e.g., pollutant removal rate, effluent compliance rate), treatment cost (e.g., reagent cost per ton of water, operation and maintenance cost), sludge production rate, and ecological safety indicators (e.g., reagent residual metal concentration).

[0051] Step S250: Establish a fast query index for partition and decision knowledge, and quickly call the corresponding historical best decision knowledge in the knowledge base through partition identifiers.

[0052] In a preferred embodiment of the present invention, the partition-decision knowledge base is equipped with a dynamic update mechanism. Based on newly added water quality monitoring data, coagulation treatment effect data, and decision execution data, the decision information of each partition can be continuously supplemented and improved, the integrity and accuracy of the historical experience base can be continuously optimized, and a fast query index for partition and decision information can be established. When the system obtains real-time water quality data and completes partition matching, the corresponding historical best decision information in the knowledge base can be quickly called through the partition identifier, which greatly improves the decision efficiency and reliability of the hybrid intelligent evaluation model.

[0053] For example, standardized optical feature data from year-round monitoring of the Yellow River section in Ningxia were used to train a self-organizing map network, completing the mapping from high-dimensional data to a low-dimensional topological grid and initially presenting the clustered distribution of water quality characteristics. Subsequently, a dynamic density clustering algorithm was used to automatically identify multiple high-density core areas, ultimately delineating multiple major water quality zones and constructing a complete dynamic water quality zoning map. One zone, characterized by high humic acid content and moderate turbidity, mainly occurs in spring and autumn and is assigned a unique identifier; another zone, characterized by low fluorescence intensity and high turbidity, mainly occurs during the flood season and corresponds to a specific identifier. The system correlates each zone identifier with historical data, discovering that in the high humic acid zone, the treatment effect of compound coagulants is optimal, and relevant information is entered into the zone-decision knowledge base. Real-time monitoring data for a certain day, after processing, is accurately classified into the high humic acid zone, providing an index for subsequent model decisions.

[0054] Step S300: Based on the standardized optical dataset and water quality zoning labels, construct a corresponding hybrid intelligent evaluation model for each dynamic water quality zoning, establish a multi-objective optimization decision function, and solve for the optimal coagulant decision scheme.

[0055] Please refer to Figure 4 In this embodiment of the invention, based on standardized optical datasets and water quality zoning labels, a corresponding hybrid intelligent evaluation model is constructed for each dynamic water quality zoning, which may include:

[0056] Step S310: Based on the preset coagulation reaction mechanism (e.g., colloidal stability theory, precipitation trapping theory, adsorption bridging theory and other classic coagulation reaction mechanisms), construct the mechanism kernel of the intelligent evaluation model for mixing.

[0057] In a preferred embodiment of the invention, the core mechanism uses key physicochemical parameters corresponding to the water quality zone as input. These core input parameters include Zeta potential, alkalinity, ionic strength, turbidity, and organic matter concentration, which directly determine the coagulation effect and the required dosage of reagents. By constructing precise physicochemical equations, the core mechanisms of action in the coagulation process, such as charge neutralization, adsorption bridging, and sedimentation trapping, are clearly described. Charge neutralization is described by the correlation equation between Zeta potential and reagent dosage; adsorption bridging is modeled by combining organic matter concentration and reagent molecular structure characteristics; and sedimentation trapping is characterized by the sedimentation kinetics equation based on particle concentration and reagent hydrolysis products. Finally, by integrating the equations relating various mechanisms, the theoretical reagent dosage required to achieve the target destabilization effect (e.g., Zeta potential approaching zero and complete destabilization of colloidal particles) is calculated, providing a fundamental theoretical constraint for the subsequent optimization of the AI-driven module.

[0058] Step S320: Construct a deep neural network as the AI-driven module of the hybrid intelligent evaluation model.

[0059] In a preferred embodiment of the present invention, the AI-driven module adopts a hybrid network structure combining a multilayer perceptron and a convolutional neural network. The convolutional neural network is used to extract deep correlation information from optical feature indicators, while the multilayer perceptron is used to fit the complex nonlinear relationship between water quality parameters, process parameters, and coagulation treatment effect. The core inputs of the AI-driven module include optical feature indicators, water quality zoning labels, and real-time on-site process parameters (e.g., influent flow rate, stirring intensity, hydraulic retention time). The outputs include parameters such as coagulant dosage and predicted treatment effect.

[0060] Step S330: Use historical verification data (e.g., actual treatment effects, reagent costs, sludge yield, etc. corresponding to different water quality conditions, different coagulant types, and different dosing parameters) as supervision signals to train the hybrid intelligent evaluation model.

[0061] Step S340: Adaptive moment estimation algorithm is used to adjust the weights and bias parameters of the deep neural network and introduce a regularization mechanism. Through multiple rounds of iterative training, the calculation results of the hybrid intelligent evaluation model are optimized and corrected.

[0062] In this embodiment of the invention, the core optimization objectives of the multi-objective optimization decision function can include four categories: first, the treatment effect objective, which focuses on maximizing the removal rate of pollutants such as turbidity, organic matter, and heavy metals to ensure that the effluent quality meets the preset compliance requirements; second, the economic objective, which aims to minimize the cost of chemicals per ton of water and the cost of sludge treatment to reduce the operation and maintenance costs of the water plant; third, the ecological safety objective, which controls the concentration of residual metals in the effluent and the degree of harmlessness of sludge to avoid secondary pollution; and fourth, the process adaptation objective, which ensures that the dosing scheme is adapted to the capabilities of the on-site process equipment to avoid process fluctuations caused by excessive or insufficient dosing of chemicals. A non-dominated sorting genetic algorithm is used to solve the multi-objective comprehensive evaluation function, generating a set of Pareto optimal solutions. This set of solutions covers the optimal decision schemes under different objective weights, with each scheme corresponding to different combinations of treatment effect and cost, providing decision-makers with flexible selection space. Simultaneously, combined with a zoning-decision knowledge base, the Pareto optimal solution set is screened, prioritizing the recommendation of decision schemes that align with the historical best treatment experience of the water quality zone and are adapted to the on-site process conditions.

[0063] For example, a dedicated intelligent evaluation model is constructed for water quality zones with high humic acid and moderate turbidity. The core mechanism is based on classical coagulation theory, using key parameters such as zeta potential and alkalinity of the zone as input to calculate the theoretical reagent dosage required for colloid destabilization. The AI-driven module uses the zone's optical characteristics and zone labels as input, combined with historical data on the treatment effects and costs of different coagulants, to learn the nonlinear relationship between humic acid content and coagulant compatibility. A multi-objective comprehensive evaluation function is constructed to optimize the balance between pollutant removal rate and treatment cost, and a Pareto optimal solution set is generated through a non-dominated sorting genetic algorithm. Combined with a zone-decision knowledge base for selection, the final output is a decision scheme with a compound coagulant as the preferred reagent, while also providing a reagent ratio range suitable for the zone, balancing treatment effectiveness and economy.

[0064] Step S400: Using digital twin technology, construct a lightweight digital twin model corresponding to the target water plant area, simulate and extrapolate the optimal coagulant decision scheme, and compare the extrapolation results with the preset water quality compliance requirements to perform feedforward optimization and strategy calibration.

[0065] Please refer to Figure 5 In this embodiment of the invention, a lightweight digital twin model corresponding to the target water plant area is constructed using digital twin technology to simulate and deduce the optimal coagulant decision scheme, which may include:

[0066] Step S410: The lightweight digital twin model covers the process units of the entire coagulation process, ensuring that the lightweight digital twin model is consistent with the physical process units.

[0067] In a preferred embodiment of the present invention, the process units of the entire coagulation treatment process include rapid mixing tanks, flocculation tanks, sedimentation tanks, etc. The modeling of each process unit is strictly based on the actual tank structure parameters (e.g., tank length, tank width, water depth) and process operation parameters (e.g., stirring intensity, hydraulic residence time, water flow velocity). The lightweight digital twin model is based on the laws of conservation of mass, conservation of momentum, and coagulation reaction kinetics to construct the core calculation equations. The rapid mixing tank focuses on simulating the rapid and uniform mixing of the agent and the water, the flocculation tank simulates the dynamic process of floc formation and growth, and the sedimentation tank simulates the floc settling and separation effect. The lightweight digital twin model is associated with the collected real-time water quality data and coagulant decision scheme parameters to achieve synchronous mapping between the virtual model and the actual process. That is, when the physical process parameters and water quality data change, the virtual model can be updated synchronously in real time to accurately restore the actual working conditions on site.

[0068] Step S420: Load the optimal coagulant decision scheme into the lightweight digital twin model, perform millisecond-level full-process simulation, and restore the dynamic process of coagulation treatment.

[0069] In a preferred embodiment of the present invention, the optimal coagulant decision scheme (e.g., including agent type, dosage, dosing rate, and dosing point) is combined with real-time process parameters (e.g., influent flow rate, water temperature, and stirring intensity) and loaded into the digital twin system to conduct millisecond-level full-process simulation, fully restoring the dynamic process of coagulation treatment. During the simulation, the system updates the water quality parameters and floc characteristic parameters (e.g., floc particle size, density, and settling velocity) of each process unit at a millisecond-level frequency, accurately simulating the complete process after agent addition, from agent diffusion and colloid destabilization in the rapid mixing stage, to floc formation and growth in the flocculation stage, and finally to floc settling and solid-liquid separation in the sedimentation tank stage.

[0070] Step S430: During the simulation, predict the water quality index values ​​of each target node, compare them with the preset water quality compliance requirements, and check for process compatibility issues in the optimal coagulant decision scheme.

[0071] In a preferred embodiment of the present invention, the water quality indicators at each key node are predicted in real time, with a focus on predicting core indicators such as effluent turbidity, chemical oxygen demand, and residual metal concentration of reagents, and comparing them in real time with preset water quality compliance requirements. Through full-process simulation, potential process compatibility issues in the decision-making scheme can be comprehensively investigated, such as insufficient reagent dosage leading to substandard pollutant removal, excessive dosage leading to cost waste and excessively high sludge production, and unreasonable dosing rate leading to uneven mixing.

[0072] In this embodiment of the invention, when comparing the simulation results with the preset water quality compliance requirements for feedforward optimization and strategy calibration, if the prediction results show that the water quality does not meet the preset standards or there are obvious process adaptation problems, the difference between the prediction results and the compliance requirements (e.g., pollutant removal rate gap, poor floc settling effect, etc.) is used as a feedforward signal and fed back to the hybrid intelligent evaluation model in the third step in real time. The model adjusts its internal parameters (e.g., theoretical calculation coefficients of the mechanism kernel, weight parameters of the AI-driven module) based on the difference information, re-optimizes the coagulant addition strategy, and generates a corrected decision scheme. If the prediction results meet the water quality compliance requirements, the optimization space of reagent dosage and treatment cost is further analyzed. By fine-tuning the addition parameters (e.g., optimizing the reagent ratio, adjusting the dosing rate), the treatment cost is reduced and the sludge production rate is reduced while ensuring the treatment effect, thereby improving the economy and stability of the scheme. After multiple rounds of simulation-comparison-feedback-correction cycle verification and calibration, the potential hidden dangers of the decision scheme are completely eliminated, and the final optimized decision scheme that can be directly applied to the actual process is output, ensuring the safety, effectiveness, and economy of the scheme.

[0073] For example, a compound coagulant dosing scheme for a high-humic acid zone was simulated and validated using a digital twin system. The digital twin model recreated the structure and process parameters of the actual water plant's rapid mixing tank, flocculation tank, and sedimentation tank. Inputting parameters such as real-time influent flow rate, it simulated the complete process after coagulant dosing in milliseconds. The simulation revealed that if the original dosing scheme was followed, fluctuations in influent flow rate would result in insufficient flocculation time, potentially causing the effluent turbidity to fail to meet standards. The system used this discrepancy as a feedforward signal to the intelligent evaluation model, which adjusted the compound coagulant ratio and dosing rhythm, generating a revised scheme. This revised scheme was then re-simulated using the digital twin system, predicting that the effluent turbidity would meet the standards and that the coagulant dosage was more reasonable. Ultimately, this revised scheme was determined to be the optimal and feasible decision.

[0074] Step S500: Using federated learning technology, all target water plant areas applying the iron-aluminum coagulant suitability evaluation method are treated as independent nodes. Each independent node deploys a corresponding hybrid intelligent evaluation model, performs local incremental model training, calculates the gradient information of model parameters, and realizes the collaborative update of the hybrid intelligent evaluation model.

[0075] Please refer to Figure 6 In this embodiment of the invention, performing incremental training of the local model and calculating the gradient information of the model parameters may include:

[0076] Step S510: Based on the new data generated by each node running locally, conduct incremental training of the local model.

[0077] In a preferred embodiment of the present invention, the core training data includes locally collected real-time water quality data, decision-making schemes verified by the digital twin system, actual coagulation treatment effect data, operation and maintenance cost data, etc. All of these data are stored on local servers and do not need to be uploaded to the central server, thus protecting the local data privacy and trade secrets (e.g., process parameters, treatment costs) of each water plant from the source.

[0078] Step S520: During training, retain the core mechanism of the hybrid intelligent evaluation model and incrementally optimize the parameters of the AI-driven module.

[0079] In a preferred embodiment of the present invention, the focus is on adapting to local water quality characteristics (e.g., regional specific pollutants, water quality fluctuation patterns) and process characteristics (e.g., equipment models, operating parameter ranges) to improve the local model's decision-making relevance and accuracy for its own water plant's water quality scenarios.

[0080] Step S530: Set up a corresponding training log for each node to record parameter adjustments and model performance changes during the training process.

[0081] Step S540: After incremental training is completed, the update amount of the model parameters is calculated as gradient information, and the gradient information is encrypted using a homomorphic encryption algorithm.

[0082] In a preferred embodiment of the present invention, after each node completes incremental training of the local model, it does not directly upload the original training data or complete local model parameters, but only calculates the update amount of the model parameters (i.e., gradient information). The gradient information reflects the contribution of local data to the optimization of model parameters, which can realize knowledge sharing and avoid leakage of original data. To further protect data privacy and security, a homomorphic encryption algorithm is used to encrypt the gradient information. This algorithm can complete the calculation and integration of data without decryption, effectively preventing the gradient information from being stolen, tampered with or cracked during transmission, and building a solid data privacy defense line.

[0083] Step S550: Construct a preset secure data transmission channel, adopt an encrypted transmission protocol, restrict transmission permissions, and transmit encrypted gradient information to the central aggregation server.

[0084] In a preferred embodiment of the present invention, only nodes are allowed to transmit encrypted gradient information to the central aggregation server, and unrelated devices are prohibited from accessing the server, thereby further enhancing the security of the transmission process. After encryption, each node uploads the encrypted gradient information to the central aggregation server according to a preset period, realizing a collaborative mode of "data not leaving the factory and knowledge being shared," which takes into account both the needs of data privacy protection and multi-party collaborative optimization.

[0085] In this embodiment of the invention, during the secure aggregation process, the encrypted gradient information uploaded by each node is first decrypted and verified to remove abnormal gradient data (e.g., invalid gradients caused by node equipment failure or data anomalies), ensuring that the gradient information participating in the aggregation is authentic and valid. Subsequently, a weighted average of all valid gradient information is applied to generate a global gradient update quantity that integrates water quality scenario knowledge and process experience from the entire network. The weight allocation is dynamically adjusted based on the data quality and water quality scenario representativeness of each node, improving the adaptability of the global model. Based on the global gradient update quantity, the parameters of the initial global hybrid intelligent evaluation model are optimized and adjusted to generate a global update model with stronger generalization ability and adaptability to more water quality scenarios. Finally, the updated global model parameters are encrypted using a homomorphic encryption algorithm and distributed to each federated learning node. Each node receives the updated parameters and synchronously updates its local model parameters, completing one collaborative update cycle. This cycle can be continuously carried out according to a preset period, enabling continuous iterative upgrades of the model and continuously improving the decision-making accuracy of the model under various special water quality scenarios (e.g., low temperature and low turbidity, high pollution load).

[0086] For example, multiple water plants deploying this system form a federated learning node cluster, initially all using a hybrid intelligent evaluation model trained on data from the Ningxia section of the Yellow River. One of these water plants, encountering a special water quality scenario of low temperature and low turbidity during the winter's low-temperature period, explored a highly efficient coagulant dosing strategy adapted to this scenario through local experiments and system operation, generating a large amount of dedicated local data. This water plant, acting as a node, used this data to incrementally train its local model, calculating the model parameter gradients and encrypting and uploading them to the central aggregation server. The central server, through a secure aggregation algorithm, integrated the gradient information of this node with other nodes, generating a global gradient update, optimizing the global model parameters, and incorporating the knowledge of handling low-temperature and low-turbidity scenarios into the global model. The updated global model is then distributed to all nodes. Subsequently, when other water plants encounter similar low-temperature and low-turbidity water quality, the system can directly output accurate coagulant decision-making schemes, achieving knowledge sharing and collaborative optimization.

[0087] Step S600: Establish a communication connection with the controller of the target water plant area through the industrial IoT interface, connect the optimal coagulant decision scheme with the actual process execution, and deploy an online water quality monitor at the end of the water coagulation and sedimentation process to monitor changes in the effluent water quality, and carry out dynamic water quality early warning and closed-loop feedback.

[0088] In this embodiment of the invention, an industrial-grade encryption protocol is used during communication to ensure the secure transmission of decision-making instructions (e.g., reagent type, dosage, and dosing rate), preventing tampering or loss of instructions. To ensure the stability of the coagulation process and avoid fluctuations in treatment effects caused by sudden jumps in reagent dosage (e.g., floc breakage, insufficient destabilization), the controller employs a smooth transition control algorithm. It compares the target dosage output in the fourth step with the current actual dosage, calculates a reasonable adjustment gradient, and transforms the change in reagent dosage into a continuous and smooth adjustment curve, gradually adjusting to the target dosage. During the adjustment process, it adapts in real-time to fluctuations in influent flow and water quality, dynamically fine-tuning the dosing rate. Simultaneously, the operating status of the dosing pump is monitored in real-time to ensure stable dosing, avoid dosing deviations caused by equipment failure, ensure smooth operation of rapid mixing, flocculation reaction, and other process stages, improve the stability and reliability of the coagulation treatment effect, and ensure that the decision-making scheme can fully function.

[0089] Please refer to Figure 7 In this embodiment of the invention, deploying an online water quality monitor to monitor changes in effluent water quality and perform dynamic early warning and closed-loop feedback for water quality can include:

[0090] Step S610: Based on the fluctuation pattern of influent water quality, historical treatment data and preset water quality compliance requirements, construct an adaptive and dynamically adjustable early warning threshold system.

[0091] In a preferred embodiment of the invention, multiple high-precision online water quality monitors are deployed at the end of the coagulation and sedimentation process, i.e., the effluent channel of the sedimentation tank, to construct a comprehensive, high-frequency real-time monitoring network for effluent water quality. The core indicators monitored by the monitors include effluent turbidity, chemical oxygen demand (COD), residual metal concentration of chemicals, and pH. The monitoring frequency is consistent with the front-end data acquisition frequency (no less than minutes) to ensure the timeliness and continuity of the data. The data collected by the monitors is transmitted in real-time to the system data processing module via an industrial IoT interface, and is synchronously compared with the prediction results of the digital twin system to calculate the deviation value and accurately evaluate the applicability and effectiveness of the coagulant dosing decision scheme. Simultaneously, through trend analysis algorithms, the dynamic trends of effluent water quality are captured, and early signs of water quality anomalies (e.g., continuously rising effluent turbidity, excessive chemical residues) are identified, providing comprehensive and real-time data support for subsequent dynamic early warning triggering and feedback optimization.

[0092] Step S620: When the deviation between the monitored effluent water quality indicators and the predicted values ​​of the lightweight digital twin model continues to exceed the dynamic early warning threshold, or when the effluent indicators reach the water quality compliance limit, a graded early warning is triggered, and the early warning information is pushed to the operation and maintenance personnel.

[0093] In a preferred embodiment of the present invention, a dynamic early warning threshold system is constructed based on the fluctuation pattern of influent water quality, historical treatment data, and preset water quality compliance requirements. The threshold is not fixed and can be adaptively adjusted according to the type and fluctuation range of real-time influent water quality, combined with historical data in the zoning-decision knowledge base, to avoid the problems of delayed warnings, false alarms, or missed alarms caused by fixed thresholds. When the deviation between the real-time monitored effluent indicators and the model prediction values ​​continuously exceeds the dynamic early warning threshold, or when the effluent indicators reach the compliance limit, the system immediately triggers a graded early warning (e.g., general early warning, emergency early warning) and automatically pushes the early warning information to relevant operation and maintenance personnel to remind them to investigate the problem in a timely manner.

[0094] Step S630: After the water quality dynamic early warning is triggered, it automatically determines whether the water quality or process unit has changed, and feeds back the monitoring data and early warning information to the beginning of the evaluation process to conduct a new round of applicability evaluation.

[0095] In a preferred embodiment of the present invention, after the early warning is triggered, the system automatically determines whether there has been a significant change in water quality or process (e.g., sudden pollution, equipment failure), and feeds back real-time monitoring data and early warning information to the system startup, driving a new round of multi-dimensional water quality fingerprint data collection, preprocessing, water quality zoning, mixed intelligent evaluation model decision-making and digital twin verification optimization, quickly adjusting the coagulant dosing strategy, generating new execution instructions and sending them to the dosing system.

[0096] For example, the system sends the optimized compound coagulant dosing plan to the programmable logic controller (PLC) of the dosing system via an industrial IoT interface. The PLC uses a smooth transition control algorithm to gradually adjust the dosing amount from the dosing pump, ensuring process stability. A high-precision monitor in the sedimentation tank effluent channel tracks the effluent turbidity in real time. Initially, the monitoring results are stable within the acceptable range and have a small deviation from the model prediction, leading the system to determine normal operation. After a period of time, affected by upstream water flow, the influent water quality undergoes a sudden change. Monitoring shows a continuous increase in turbidity, and the deviation from the predicted value exceeds the dynamic warning threshold, triggering an immediate warning. Following the warning, the system feeds back the real-time monitoring data to the front end, re-acquires and preprocesses data, adjusts water quality zoning, re-optimizes the dosing strategy using the model, and issues new execution commands to ensure the effluent water quality returns to acceptable levels.

[0097] Step S700: Establish a routine operation and maintenance mechanism, perform accuracy calibration of multimodal sensors and monitoring instruments according to the preset cycle, and conduct a full-process data review and analysis of the iron-aluminum coagulant applicability evaluation system. Combine the review and analysis results with the federated learning collaborative update results to perform iterative optimization of all modules of the system.

[0098] In this embodiment of the invention, daily inspections can be conducted on hardware such as the sensor network, dosing equipment, and online monitoring instruments to troubleshoot equipment malfunctions (e.g., sensor failure, dosing pump jamming) and wiring connection problems, and to promptly address any abnormal equipment conditions. Weekly cleaning and maintenance can be performed to remove dirt from sensor probes and residual chemicals from dosing pipelines to prevent impurities from affecting equipment operating accuracy. Monthly comprehensive maintenance can be carried out to replace aging components (e.g., sensor probes, pipeline seals) to ensure long-term stable operation of the equipment. Regular precision calibration of sensors and monitoring instruments can be performed using standard sample comparison to correct measurement deviations, eliminate errors caused by equipment wear, and ensure the accuracy of data acquisition and monitoring.

[0099] In this embodiment of the invention, the review process focuses on analyzing three types of problems: first, water quality zoning deviation, investigating factors such as the adaptability of the zoning algorithm and the recognizability of feature data, and analyzing the model decision deviation caused by zoning deviation; second, the difference between model decisions and actual execution results, comparing model predictions with actual processing effects, and optimizing model parameter adaptability; and third, the rationality of early warning triggering, analyzing the causes of false alarms and missed alarms, and adjusting the logic for setting early warning thresholds. Simultaneously, the system adaptation experience for different water quality scenarios (e.g., flood season, low-temperature winter period) is summarized, the optimal operation and maintenance scheme and decision-making strategy are identified, the optimization space of each module of the system is accurately identified, a complete review report is generated, and the root causes of problems, improvement directions, and implementation plans are clarified.

[0100] In this embodiment of the invention, the system undergoes iterative optimization across all modules by combining monthly and quarterly review analysis results with the collaborative update results of federated learning, thereby achieving continuous improvement in system performance. The data acquisition and preprocessing module optimizes the feature extraction algorithm and adjusts the parameters of wavelet transform and principal component analysis to improve data identification in high-interference water quality scenarios and reduce invalid data. The water quality zoning module optimizes the parameters of the self-organizing mapping network and dynamic density clustering algorithm to improve zoning accuracy and dynamic adaptability, reducing zoning bias. The hybrid intelligent evaluation model optimizes the physicochemical equation coefficients of the mechanism kernel, the network structure and parameters of the AI-driven module, and combines the global knowledge of federated learning to improve the model's decision-making accuracy and generalization ability. The digital twin system optimizes the process unit model parameters to improve the accuracy and speed of simulation and better adapt to process fluctuations. The early warning module optimizes the dynamic threshold setting logic, adjusts the threshold adaptive mechanism based on early warning issues discovered during reviews, and reduces false alarms and false negatives. After the iteration is completed, the optimization effect is verified through small-scale field tests. The system performance before and after the iteration is compared (e.g., decision accuracy, processing effect stability, and early warning rationality) to ensure that the optimized system performance is significantly improved. At the same time, the optimization results are synchronized to each federated learning node to achieve collaborative upgrade of the entire network system.

[0101] For example, a monthly operation and maintenance mechanism was established. Each month, the sensor network, water treatment plant dosing equipment, and effluent monitoring instruments at the monitoring sections of the Yellow River in Ningxia were inspected and calibrated. Aging fluorescence spectral sensors were replaced, and the accuracy of turbidimeters was calibrated to ensure accurate data acquisition. Simultaneously, the digital twin system and federated learning nodes were inspected to identify potential communication link vulnerabilities. Monthly full-process data reviews revealed that certain water quality zones were prone to zoning bias during the flood season, leading to model decision-making errors. Additionally, the early warning module experienced a small number of false alarms. Based on the review results, the dynamic density clustering algorithm parameters were optimized to improve zoning accuracy; the early warning threshold setting logic was adjusted, and the threshold adaptive mechanism was optimized in conjunction with the water quality fluctuation patterns during the flood season; and the global model parameters updated by federated learning were integrated to optimize the hybrid intelligent evaluation model. After iteration, field tests verified that the zoning bias rate and early warning false alarm rate were significantly reduced, and the model decision-making accuracy was improved, ensuring stable system operation even during the flood season.

[0102] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0107] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0108] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), 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 technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0109] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0110] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning, characterized in that, The method for evaluating the suitability of the iron-aluminum coagulant includes: Based on multimodal sensors deployed in the target water plant area, multidimensional data of the water body are collected, and standardized optical datasets are generated through purification, feature extraction and enhancement processing. A self-organizing mapping network is used to transform high-dimensional data into low-dimensional space in a standardized optical dataset. A dynamic density clustering algorithm is used to delineate the boundaries of water quality zones and associate historical decision knowledge with each water quality zone to generate water quality zone labels and zone maps. Based on standardized optical datasets and water quality zoning labels, a corresponding hybrid intelligent evaluation model is constructed for each dynamic water quality zoning, and a multi-objective optimization decision function is established to solve for the optimal coagulant decision scheme. Using digital twin technology, a lightweight digital twin model corresponding to the target water plant area is constructed to simulate and extrapolate the optimal coagulant decision scheme. The simulation results are then compared with the preset water quality compliance requirements to perform feedforward optimization and strategy calibration.

2. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 1, characterized in that, The collected multi-dimensional water body data, through purification, feature extraction, and enhancement processing, generates a standardized optical dataset, including: An adaptive noise reduction algorithm is used to suppress noise in multi-dimensional water body data, eliminating noise generated by equipment operation and environmental interference, while retaining the detailed features of the effective spectrum. We performed internal filtration effect correction on the multi-dimensional water data after noise suppression, constructed a quantitative correlation between water concentration and fluorescence intensity attenuation, and performed point-by-point correction on the spectral signal. Raman scattering is subtracted from the point-by-point corrected spectral signal, and the baseline correction method is used to remove Raman scattering interference, highlighting the inherent optical characteristics of the water body, and obtaining the purified water spectral data. Wavelet transform was used to perform time-frequency analysis on the spectral data of the purified water body, and principal component analysis was used to reduce the dimensionality of the high-dimensional wavelet feature data, extract the target optical feature indicators, and generate a standardized optical dataset.

3. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 1, characterized in that, The method employs a dynamic density clustering algorithm to delineate water quality zoning boundaries and associates historical decision-making knowledge with each water quality zoning zone, including: Based on the clustering results of the low-dimensional topological grid generated by the self-organizing mapping network, a dynamic density clustering algorithm is adopted, and a density threshold is set to identify the high-density core region and low-density boundary region of water quality characteristic data points in the low-dimensional topological grid. When the quality of the incoming water changes, the distribution density of the water quality characteristic data points changes accordingly. The dynamic density clustering algorithm automatically identifies the changes and dynamically adjusts the range and number of partition boundaries. Identified water quality anomalies are individually marked to ensure that the zoning results accurately reflect the water quality status. Each identified stable water quality zone is assigned a corresponding zone identifier, and a dynamically updated zone-decision knowledge base is constructed, linking each zone identifier with the corresponding historical decision knowledge. Establish a fast query index for partition and decision knowledge, and quickly retrieve the corresponding historical best decision knowledge from the knowledge base through partition identifiers.

4. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 1, characterized in that, The method, based on standardized optical datasets and water quality zoning labels, constructs a corresponding hybrid intelligent evaluation model for each dynamic water quality zoning, including: Based on the pre-defined coagulation reaction mechanism, the core mechanism of the hybrid intelligent evaluation model is constructed. Construct a deep neural network as the AI-driven module of a hybrid intelligent evaluation model; The hybrid intelligent evaluation model is trained using historical validation data as a supervision signal; An adaptive moment estimation algorithm is adopted to adjust the weights and bias parameters of the deep neural network and introduce a regularization mechanism. Through multiple rounds of iterative training, the calculation results of the hybrid intelligent evaluation model are optimized and corrected.

5. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 1, characterized in that, The process involves using digital twin technology to construct a lightweight digital twin model corresponding to the target water plant area, and simulating and extrapolating the optimal coagulant decision-making scheme, including: The lightweight digital twin model covers all process units of the coagulation process, ensuring that the lightweight digital twin model is consistent with the physical process units. The optimal coagulant decision scheme is loaded into a lightweight digital twin model to perform millisecond-level full-process simulation and restore the dynamic process of coagulation treatment. During the simulation, the predicted water quality index values ​​for each target node are compared with the preset water quality compliance requirements to identify process compatibility issues in the optimal coagulant decision-making scheme.

6. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 1, characterized in that, The method for evaluating the suitability of iron-aluminum coagulants also includes: By employing federated learning technology, all target water plant areas applying the iron-aluminum coagulant suitability evaluation method are treated as independent nodes. Each independent node deploys a corresponding hybrid intelligent evaluation model, performs local incremental model training, calculates the gradient information of model parameters, and achieves collaborative updates of the hybrid intelligent evaluation model.

7. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 6, characterized in that, The step of performing incremental training of the local model and calculating the gradient information of the model parameters includes: Based on the new data generated by each node running locally, conduct incremental training of the local model. During training, the core mechanism of the hybrid intelligent evaluation model is preserved, and the parameters of the AI-driven module are incrementally optimized. Each node is configured with a corresponding training log to record parameter adjustments and model performance changes during the training process; After incremental training is completed, the update amount of the model parameters is calculated as gradient information, and the gradient information is encrypted using a homomorphic encryption algorithm. A pre-defined secure data transmission channel is established, an encrypted transmission protocol is adopted, transmission permissions are restricted, and each node transmits encrypted gradient information to the central aggregation server.

8. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 6, characterized in that, The method for evaluating the suitability of iron-aluminum coagulants also includes: Through the industrial IoT interface, a communication connection is established with the controller of the target water plant area to connect the optimal coagulant decision scheme with the actual process execution. At the end of the water coagulation and sedimentation process, an online water quality monitoring instrument is deployed to monitor changes in the effluent water quality and to carry out dynamic water quality early warning and closed-loop feedback.

9. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 8, characterized in that, The deployed online water quality monitoring instrument monitors changes in effluent water quality, providing dynamic early warning and closed-loop feedback, including: Based on the fluctuation patterns of influent water quality, historical treatment data, and preset water quality compliance requirements, an adaptive and dynamically adjustable early warning threshold system is constructed. When the deviation between the monitored effluent water quality indicators and the predicted values ​​of the lightweight digital twin model continues to exceed the dynamic early warning threshold, or when the effluent indicators reach the water quality compliance limit, a graded early warning will be triggered and the early warning information will be pushed to the operation and maintenance personnel. Once a dynamic water quality warning is triggered, the system automatically determines whether there have been changes in the water quality or process unit, and feeds back the monitoring data and warning information to the beginning of the evaluation process for a new round of applicability evaluation.

10. The method for evaluating the suitability of iron-aluminum coagulants based on water quality zoning according to claim 8, characterized in that, The method for evaluating the suitability of iron-aluminum coagulants also includes: Establish a routine operation and maintenance mechanism, perform accuracy calibration of multimodal sensors and monitoring instruments according to a preset cycle, and conduct a full-process data review and analysis of the iron-aluminum coagulant applicability evaluation system. Combine the review and analysis results with the federated learning collaborative update results to carry out iterative optimization of all modules of the system.

Images