Fluoride removal method and system for fluorine-containing wastewater by crystallization driven by digital twin

Abstract

The application provides a digital twin driven fluorine-containing wastewater crystallization defluorination intelligent control method and system, and relates to the technical field of fluorine-containing wastewater treatment. The fluorine-containing wastewater crystallization defluorination intelligent control method adopts a multi-model collaborative architecture, obtains standardized multi-source time series data, combines the target crystallization kinetics precipitation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model, calculates the target control parameter by using the reinforcement learning decision model, realizes dynamic control of the kinetics characteristics of the fluorine-containing wastewater crystallization defluorination process, improves the control precision of the system crystallization kinetics precipitation rate, effectively overcomes the defects of kinetics inaccuracy and low control precision of the system crystallization kinetics precipitation rate caused by the traditional manual experience setting of fixed control parameters, ensures that the effluent fluorine ion concentration is stable and up to standard, reduces the consumption of calcium salt reagent and the amount of sludge generated, and significantly improves the system operation stability, reliability and economy.

Description

Technical Field

[0001] This application relates to the field of fluoride-containing wastewater treatment technology, and in particular to a digital twin-driven intelligent control method and system for crystallization-based fluoride removal from fluoride-containing wastewater. Background Technology

[0002] Fluoride-containing wastewater is widely generated in industrial production processes such as chemical, metallurgical, electroplating, and semiconductor manufacturing. It is characterized by high fluoride ion concentrations, strong toxicity, and difficulty in biodegradation. Direct discharge of such wastewater without effective treatment poses a dual threat to the ecological environment and human health: firstly, fluoride ions pollute soil and water bodies, destroying aquatic habitats; secondly, fluoride ions can enter the human body through drinking water and crops, and long-term ingestion can easily lead to diseases such as dental fluorosis and skeletal fluorosis. Therefore, the compliant treatment of fluoride-containing wastewater has become a rigid requirement in the field of industrial environmental protection, especially given increasingly stringent environmental standards, as traditional treatment technologies are no longer sufficient to meet the requirements for efficient, stable, and low-consumption treatment.

[0003] Currently, crystallization for fluoride removal has become the mainstream technology for the deep treatment of fluoride-containing wastewater due to its advantages such as high fluoride ion removal efficiency, low sludge production, moderate operating costs, simple operation, and the ability to recover fluoride resources. Its core principle is to add calcium salts and other agents to the fluoride-containing wastewater, causing fluoride ions to react with calcium ions to form insoluble calcium fluoride crystals, which are then removed efficiently through solid-liquid separation. However, current crystallization systems for fluoride removal in wastewater generally rely on manual experience to set fixed control parameters, such as preset operating conditions for calcium salt dosage, pH value, and stirring intensity based on historical system operating data.

[0004] However, the above-mentioned control method, which relies on human experience to set fixed control parameters, will lead to serious inaccuracies in the kinetic characteristics of the crystallization and defluorination process of fluoride-containing wastewater, resulting in low accuracy in controlling the precipitation rate of the system's crystallization kinetics. Summary of the Invention

[0005] This application provides a digital twin-driven intelligent control method and system for defluoridation of fluoride-containing wastewater crystallization, which solves the problem that the control method in related technologies, which relies on manual experience to set fixed control parameters, leads to serious inaccuracies in the kinetic characteristics of the defluoridation process of fluoride-containing wastewater crystallization, resulting in low accuracy in controlling the precipitation rate of the crystallization kinetics of the system.

[0006] In a first aspect, this application provides a digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization, comprising: acquiring standardized multi-source time-series data of the defluoridation process of fluoride-containing wastewater crystallization within the current time window; inputting the standardized multi-source time-series data into a digital twin estimation model and a Long Short-Term Memory Network (LSTM) prediction model respectively to obtain the target crystallization kinetic sedimentation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model; inputting the target crystallization kinetic sedimentation rate and the predicted effluent index into a reinforcement learning decision model to calculate the control parameters, thereby obtaining the target control parameters output by the reinforcement learning decision model; and controlling the defluoridation process of fluoride-containing wastewater crystallization within the next time window adjacent to the current time window according to the target control parameters.

[0007] In one possible implementation, the digital twin estimation model includes an ion solubility product thermodynamic model, a crystallization kinetic model, and a time-series data correction model. Standardized multi-source time-series data includes the crystallization reaction temperature. The standardized multi-source time-series data is input into the digital twin estimation model to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model. This includes: calculating the calcium fluoride solubility product corresponding to the crystallization reaction temperature based on the ion solubility product thermodynamic model; calculating the simulated crystallization kinetic precipitation rate based on the standardized multi-source time-series data and the calcium fluoride solubility product based on the crystallization kinetic model; and correcting the simulated crystallization kinetic precipitation rate based on the time-series data correction model to obtain the target crystallization kinetic precipitation rate.

[0008] In one possible implementation, the standardized multi-source time-series data also includes influent parameters of fluoride-containing wastewater, calcium salt dosage parameters, crystallization reaction process parameters, and the current crystallization kinetic precipitation rate. Based on the crystallization kinetic model, the simulated crystallization kinetic precipitation rate is calculated using the standardized multi-source time-series data and the calcium fluoride solubility product. This includes: calculating the current calcium ion concentration and current fluoride ion concentration in the crystallization reaction tank based on the crystallization kinetic model, the influent parameters of fluoride-containing wastewater, the calcium salt dosage parameters, and the current crystallization kinetic precipitation rate; calculating the supersaturation ratio based on the current calcium ion concentration, the current fluoride ion concentration, and the calcium fluoride solubility product; and calculating the simulated crystallization kinetic precipitation rate based on the supersaturation ratio, the crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetic model.

[0009] In one possible implementation, the intelligent control method for defluoridation of fluoride-containing wastewater crystallization further includes: calculating the crystallization kinetic sedimentation rate deviation between the current crystallization kinetic sedimentation rate and the simulated crystallization kinetic sedimentation rate; using a PID control algorithm to calculate the compensation adjustment amount of the current growth rate compensation constant based on the crystallization kinetic sedimentation rate deviation; and accumulating the compensation adjustment amount to the current growth rate compensation constant to obtain the target growth rate compensation constant.

[0010] In one possible implementation, the simulated crystallization kinetics precipitation rate is calculated based on the supersaturation ratio, crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetics model. This includes iteratively calculating the simulated crystallization kinetics precipitation rate based on the supersaturation ratio, crystallization reaction process parameters, and the target growth rate compensation constant.

[0011] In one possible implementation, the target crystallization kinetic sedimentation rate and the predicted effluent index are input into a reinforcement learning decision model to calculate the control parameters, thereby obtaining the target control parameters output by the reinforcement learning decision model. This includes: inputting the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model, taking the target crystallization kinetic sedimentation rate as the core control objective and combining the compliance constraints of the predicted effluent index to output the target control parameters.

[0012] In one possible implementation, the target control parameters include the Ca / F molar ratio, pH setpoint, seed dosage, stirring intensity, and the number of microchannel reaction columns opened. The method further includes: determining the number of microchannel reaction columns opened and their spatial distribution pattern based on the target crystallization kinetic precipitation rate; the number of microchannel reaction columns opened is determined based on the total number of microchannel reaction columns, and the opened microchannel reaction columns are staggered at intervals; and determining the target opening degree of the seed addition control valve and the corresponding seed dosage based on the target crystallization kinetic precipitation rate, wherein the opening degree of the seed addition control valve is positively correlated with the seed dosage.

[0013] In one possible implementation, the LSTM prediction model is trained as follows: First, a first multi-source time-series training dataset is acquired, which is a preprocessed, standardized historical multi-source time-series dataset; the first multi-source time-series training dataset is input into the LSTM prediction model to be trained to predict water discharge indicators, resulting in predicted water discharge indicator values ​​output by the LSTM prediction model; the mean square error between the predicted water discharge indicator values ​​and the true values ​​of the water discharge indicators in the first multi-source time-series training dataset is used as the loss function to iteratively optimize the model parameters of the LSTM prediction model to be trained until the loss function converges, thus obtaining the LSTM prediction model.

[0014] In one possible implementation, the reinforcement learning decision model is trained as follows: acquiring second multi-source time-series training data; inputting the second multi-source time-series training data into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model to be trained; using the multi-weighted reward function as the optimization objective, iteratively optimizing the model parameters of the reinforcement learning decision model to be trained based on the control parameters until the control parameters output by the reinforcement learning decision model to be trained can stabilize the score of the multi-weighted reward function within a preset score range, thus obtaining the reinforcement learning decision model.

[0015] In one possible implementation, the multi-weighted reward function includes weights for effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability; the total score for these weights is 100 points.

[0016] In one possible implementation, standardized multi-source time-series data of the fluoride-containing wastewater crystallization defluorination process running within the current time window are obtained, including: obtaining initial multi-source time-series data of the fluoride-containing wastewater crystallization defluorination process running within the current time window; preprocessing the initial multi-source time-series data to obtain standardized multi-source time-series data, the preprocessing including outlier removal, missing value completion, data normalization, and multi-parameter time-series alignment.

[0017] In one possible implementation, the intelligent control method for defluoridation crystallization in fluoride-containing wastewater further includes: acquiring system feedback operation data of the defluoridation crystallization process in fluoride-containing wastewater within the next time window; and correcting the target control parameters based on the system feedback operation data and preset standard operation data.

[0018] In one possible implementation, the intelligent control method for defluoridation of fluoride-containing wastewater crystallization further includes: identifying system operation faults based on a fault identification model and standardized multi-source time-series data; and in response to the identification of an operation fault in the system, triggering a preset emergency control plan and alarm prompts according to the fault type corresponding to the operation fault.

[0019] Secondly, this application provides a digital twin-driven intelligent control device for defluoridation of fluoride-containing wastewater crystallization, comprising:

[0020] The acquisition module is used to acquire standardized multi-source time-series data of the crystallization and defluorination process of fluoride-containing wastewater within the current time window;

[0021] The first processing module is used to input standardized multi-source time series data into the digital twin estimation model and the LSTM prediction model respectively, to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model.

[0022] The second processing module inputs the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model to calculate the control parameters, and obtains the target control parameters output by the reinforcement learning decision model.

[0023] The control module is used to control the crystallization and defluorination process of fluoride-containing wastewater in the next time window adjacent to the current time window, based on the target adjustment parameters.

[0024] In one possible implementation, the digital twin estimation model includes an ion solubility product thermodynamic model, a crystallization kinetic model, and a time-series data correction model. The standardized multi-source time-series data includes the crystallization reaction temperature. The first processing module is specifically used to: calculate the calcium fluoride solubility product corresponding to the crystallization reaction temperature based on the ion solubility product thermodynamic model; calculate the simulated crystallization kinetic precipitation rate based on the crystallization kinetic model, the standardized multi-source time-series data, and the calcium fluoride solubility product; and correct the simulated crystallization kinetic precipitation rate based on the time-series data correction model to obtain the target crystallization kinetic precipitation rate.

[0025] In one possible implementation, the standardized multi-source time-series data also includes influent parameters of fluoride-containing wastewater, calcium salt dosing parameters, crystallization reaction process parameters, and the current crystallization kinetic sedimentation rate. The first processing module is further used to: calculate the current calcium ion concentration and current fluoride ion concentration in the crystallization reaction tank based on the crystallization kinetic model, according to the influent parameters of fluoride-containing wastewater, calcium salt dosing parameters, and the current crystallization kinetic sedimentation rate; calculate the supersaturation ratio based on the current calcium ion concentration, current fluoride ion concentration, and calcium fluoride solubility product; and calculate the simulated crystallization kinetic sedimentation rate based on the supersaturation ratio, crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetic model.

[0026] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a compensation module (not shown). This compensation module is used to: calculate the crystallization kinetic sedimentation rate deviation between the current crystallization kinetic sedimentation rate and the simulated crystallization kinetic sedimentation rate; use a PID control algorithm to calculate the compensation adjustment amount of the current growth rate compensation constant based on the crystallization kinetic sedimentation rate deviation; and accumulate the compensation adjustment amount to the current growth rate compensation constant to obtain the target growth rate compensation constant.

[0027] In one possible implementation, the first processing module is also used to iteratively calculate the precipitation rate of the simulated crystallization kinetics based on the supersaturation ratio, crystallization reaction process parameters, and target growth rate compensation constant.

[0028] In one possible implementation, the second processing module is specifically used to: input the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model, take maintaining the target crystallization kinetic sedimentation rate within the preset crystallization kinetic sedimentation rate range as the core control objective, and output the target control parameters in combination with the compliance constraints of the predicted effluent index.

[0029] In one possible implementation, the target control parameters include the Ca / F molar ratio, pH setpoint, seed dosage, stirring intensity, and the number of microchannel reaction columns opened. The control module is also used to: determine the number of microchannel reaction columns opened and the spatial distribution pattern of the microchannel reaction columns based on the target crystallization kinetic precipitation rate, wherein the number of microchannel reaction columns opened is determined based on the total number of microchannel reaction columns, and the opened microchannel reaction columns are staggered at intervals; and determine the target opening degree of the seed addition control valve and the corresponding seed dosage based on the target crystallization kinetic precipitation rate, wherein the opening degree of the seed addition control valve is positively correlated with the seed dosage.

[0030] In one possible implementation, the LSTM prediction model is trained as follows: First, a first multi-source time-series training dataset is acquired, which is a preprocessed, standardized historical multi-source time-series dataset; the first multi-source time-series training dataset is input into the LSTM prediction model to be trained to predict water discharge indicators, resulting in predicted water discharge indicator values ​​output by the LSTM prediction model; the mean square error between the predicted water discharge indicator values ​​and the true values ​​of the water discharge indicators in the first multi-source time-series training dataset is used as the loss function to iteratively optimize the model parameters of the LSTM prediction model to be trained until the loss function converges, thus obtaining the LSTM prediction model.

[0031] In one possible implementation, the reinforcement learning decision model is trained as follows: acquiring second multi-source time-series training data; inputting the second multi-source time-series training data into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model to be trained; using the multi-weighted reward function as the optimization objective, iteratively optimizing the model parameters of the reinforcement learning decision model to be trained based on the control parameters until the control parameters output by the reinforcement learning decision model to be trained can stabilize the score of the multi-weighted reward function within a preset score range, thus obtaining the reinforcement learning decision model.

[0032] In one possible implementation, the multi-weighted reward function includes weights for effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability; the total score for these weights is 100 points.

[0033] In one possible implementation, the acquisition module is specifically used to: acquire the initial multi-source time series data of the fluoride-containing wastewater crystallization defluorination process running within the current time window; preprocess the initial multi-source time series data to obtain standardized multi-source time series data, the preprocessing including outlier removal, missing value completion, data normalization, and multi-parameter time series alignment.

[0034] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a correction module (not shown), which is used to: acquire system feedback operation data of the defluoridation process of fluoride-containing wastewater crystallization within the next time window; and correct the target control parameters based on the system feedback operation data and the preset standard operation data.

[0035] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a fault identification module (not shown). This fault identification module is used to: identify system operation faults based on a fault identification model and standardized multi-source time-series data; and in response to the identification of an operation fault in the system, trigger a preset emergency control plan and alarm prompts according to the fault type corresponding to the operation fault.

[0036] Thirdly, this application provides a digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization, including: an intelligent control unit and a seed crystal regeneration and reuse unit;

[0037] An intelligent control unit is used to execute the intelligent control method for defluoridation of fluoride-containing wastewater crystallization as provided in the first aspect above.

[0038] The seed regeneration and reuse unit is used to classify and separate the calcium fluoride crystal mixture generated by the above-mentioned intelligent control method for defluorination through crystallization of fluoride-containing wastewater, and to regenerate the seed crystals that do not meet the requirements for crystal reuse.

[0039] Fourthly, this application provides an electronic device, including: a processor and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the intelligent control method for defluoridation of fluoride-containing wastewater crystallization as provided in the first aspect above.

[0040] Fifthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the intelligent control method for defluoridation of fluoride-containing wastewater by crystallization as provided in the first aspect above.

[0041] Sixthly, this application provides a computer program product, comprising: a computer program that, when executed by a processor, implements the intelligent control method for defluoridation of fluoride-containing wastewater by crystallization as provided in the first aspect above.

[0042] This application provides a digital twin-driven intelligent control method and system for defluoridation of fluoride-containing wastewater crystallization. The method acquires standardized multi-source time-series data of the defluoridation process within the current time window. This standardized multi-source time-series data is then input into a digital twin estimation model and an LSTM prediction model to obtain the target crystallization kinetic sedimentation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model. Furthermore, the target crystallization kinetic sedimentation rate and the predicted effluent index are input into a reinforcement learning decision model to calculate control parameters, obtaining the target control parameters output by the reinforcement learning decision model. Finally, based on the target control parameters, the defluoridation process of the fluoride-containing wastewater crystallization in the next time window adjacent to the current time window is controlled. This application adopts a multi-model collaborative architecture. By acquiring standardized multi-source time-series data, combining the target crystallization kinetic sedimentation rate output by the digital twin estimation model and the predicted effluent indicators output by the LSTM prediction model, and using a reinforcement learning decision model to calculate the target control parameters, dynamic control of the kinetic characteristics of the crystallization and defluoridation process of fluoride-containing wastewater is achieved. This improves the control accuracy of the system's crystallization kinetic sedimentation rate and effectively overcomes the shortcomings of traditional manual experience-based fixed control parameter setting, which leads to kinetic inaccuracies and low control accuracy of the system's crystallization kinetic sedimentation rate. It can ensure that the effluent fluoride concentration is stably up to standard, while reducing the consumption of calcium salt reagents and the amount of sludge generated, significantly improving the system's operational stability, reliability, and economy. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] Figure 1 A schematic diagram of the structure of the digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization provided in an embodiment of this application;

[0045] Figure 2 A flowchart illustrating the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater via crystallization, as provided in this application embodiment. Figure 1 ;

[0046] Figure 3 A flowchart illustrating the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater via crystallization, as provided in this application embodiment. Figure 2 ;

[0047] Figure 4 A schematic diagram of the structure of the digital twin-driven intelligent control device for defluoridation of fluoride-containing wastewater crystallization provided in the embodiments of this application;

[0048] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0049] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0050] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0051] In related technologies, fluoride removal crystallization systems for fluoride-containing wastewater generally rely on manual experience to set fixed control parameters, such as preset operating conditions like calcium salt dosage, pH value, and stirring intensity based on historical system data. However, under this control method, manual experience cannot perceive and respond to complex and changing on-site conditions in real time, leading to serious inaccuracies in the kinetic control of the fluoride removal crystallization process. This results in low accuracy in controlling the crystallization kinetic sedimentation rate and delayed response to operating conditions. Furthermore, existing fluoride removal crystallization systems lack a closed-loop linkage mechanism between the simulated crystallization kinetic sedimentation rate and on-site control parameters, still requiring manual adjustment by operators. This not only involves high labor intensity but also increases the risk of improper calcium salt dosage, fluctuations in effluent quality, and even exceeding fluoride ion concentration limits due to human error or untimely parameter adjustments.

[0052] With increasingly stringent environmental standards and the widespread application of intelligent technologies such as big data, artificial intelligence, and digital twins in industrial water treatment, traditional control methods based on fixed parameters adjusted by human experience are no longer sufficient to meet the actual needs of precise and intelligent control in the crystallization and defluoridation process of fluoride-containing wastewater. This severely restricts the industrial-scale promotion of crystallization processes. Therefore, there is an urgent need for an intelligent control method for crystallization and defluoridation of fluoride-containing wastewater that can accurately correct the sedimentation rate of simulated crystallization kinetics, achieve collaborative sensing of multiple control parameters, and possess AI intelligent parameter optimization and closed-loop feedback control functions. This method aims to overcome many pain points in related technologies, promote the intelligent upgrade of fluoride-containing wastewater crystallization and defluoridation treatment systems, and ensure that the effluent fluoride concentration continuously and stably meets the standards.

[0053] Based on the technical problems existing in related technologies, the embodiments of this application adopt a multi-model collaborative architecture. By acquiring standardized multi-source time-series data, combining the target crystallization kinetic sedimentation rate output by the digital twin estimation model and the predicted effluent indicators output by the LSTM prediction model, and using a reinforcement learning decision model to calculate the target control parameters, dynamic control of the kinetic characteristics of the crystallization and defluoridation process of fluoride-containing wastewater is achieved. This improves the control accuracy of the system's crystallization kinetic sedimentation rate, effectively overcoming the defects of kinetic inaccuracies and low control accuracy of the system's crystallization kinetic sedimentation rate caused by traditional manual experience-based fixed control parameters. It can ensure that the effluent fluoride concentration is stably up to standard, while reducing the consumption of calcium salt reagents and the amount of sludge generated, significantly improving the system's operational stability, reliability, and economy.

[0054] The application scenarios of the embodiments of this application will be described below first.

[0055] The digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater through crystallization provided in this application is applicable to high-concentration fluoride-containing wastewater treatment scenarios generated in industrial fields such as chemical, metallurgical, electroplating, and semiconductor industries.

[0056] The following is a combination of... Figure 1 This application provides a detailed description of the digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization, as provided in the embodiments of this application.

[0057] Figure 1 This is a schematic diagram of the structure of a digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization, provided in an embodiment of this application. Figure 1 As shown, the intelligent control system for defluoridation of fluoride-containing wastewater crystallization includes an inlet unit 10, a reagent dosing unit 11, a crystallization reaction unit 12, an outlet unit 13, a seed crystal regeneration and reuse unit 14, and an intelligent control unit 15.

[0058] The core components of the inlet unit 10 include an inlet pump 101, an inlet control valve 102, an inlet fluoride ion meter 103, an inlet flow meter 104, and an inlet temperature sensor 105. The inlet pump 101 is used to transport the fluoride-containing wastewater to be treated from the raw water tank to the crystallization reaction unit 12, providing the inlet power for the entire water treatment process and achieving continuous and stable delivery of the fluoride-containing wastewater. The inlet control valve 102, installed on the inlet pipe 106, is used to adjust the inlet flow rate and control the inlet flow. It can work with the intelligent control system to realize inlet start / stop and flow throttling adjustment to adapt to different treatment loads. The inlet fluoride ion meter 103 is used to collect the inlet fluoride ion concentration in the fluoride-containing wastewater in real time. The inlet flow meter 104 is used to collect the inlet flow rate of the fluoride-containing wastewater in real time. The inlet temperature sensor 105 is used to collect the inlet temperature of the fluoride-containing wastewater in real time. The inlet temperature sensor can be an inlet thermometer. Figure 1The black arrow in the inlet water unit 10 indicates the direction of the inlet flow of fluoride-containing wastewater.

[0059] The core components of the reagent dosing unit 11 include a reagent dosing pump 111, a reagent dosing control valve 112, a reagent flow meter 113, and a reagent concentration detector 114. The reagent dosing pump 111 provides the power to transport the calcium salt reagent, quantitatively and continuously delivering the prepared calcium salt reagent to the crystallization reaction unit, achieving stable calcium salt reagent dosing. The reagent dosing control valve 112 controls the on / off state of the reagent dosing pipeline 115 and adjusts the opening of the dosing valve. By adjusting the opening of the dosing valve to match real-time operating conditions, the amount of calcium salt reagent added is precisely adjusted, enabling automated and precise reagent dispensing in conjunction with the intelligent control system. The reagent flow meter 113 is used to collect the amount of calcium salt reagent added in real-time online. The reagent concentration detector 114 is used to collect the concentration of calcium salt reagent in the reagent dosing pipeline in real-time online. Figure 1 The black arrow in the Chinese medicine dosing unit 11 indicates the dosing direction of the calcium salt medicine.

[0060] The core components of the crystallization reaction unit 12 include a stirring device 121, a main microchannel control switch 122, multiple microchannel reaction columns 123, a pH meter 124, a crystal particle size analyzer 125, a crystallization concentration detector 126, and a stirring intensity sensor 127. The stirring device 121 is used to uniformly stir the fluoride-containing wastewater and calcium salt reagent within the crystallization reaction unit, promoting thorough mixing and contact between fluoride ions and calcium ions, accelerating the nucleation and growth rate of calcium fluoride crystals, maintaining uniform concentration and temperature in the reaction system, and ensuring the stable progress of the fluoride removal crystallization reaction. The main microchannel control switch 122 is connected to the multiple microchannel reaction columns 123 and is used to control the number and mode of operation of the multiple microchannel reaction columns 123. The operation mode can adopt an intermittent or staggered mode (avoiding centralized use where one side is on and the other is off). The multiple microchannel reaction columns 123 are used to provide microchannels with... The reaction environment is limited, enhancing local flow field disturbance and mass transfer effects to provide a dedicated reaction channel for the nucleation and growth of calcium fluoride crystals. The system can adapt to different crystallization kinetics and precipitation rate conditions by controlling the number and interval distribution of activated components, achieving precise reaction control. A pH meter (124) is used to collect the pH value of the water within the crystallization reaction unit in real time. A crystal particle size analyzer (125) is used to collect the crystal particle size distribution and average crystal particle size of calcium fluoride crystals within the crystallization reaction system in real time. A crystal concentration detector (126) is used to collect the crystal concentration within the crystallization reaction unit in real time. A stirring intensity sensor (127) is used to collect the stirring intensity within the crystallization reaction unit in real time.

[0061] The core components of the effluent unit 13 include an effluent control valve 131 and an effluent fluoride ion meter 132. The effluent control valve 131 is used to control the opening and closing of the effluent pipeline and regulate the effluent flow rate of the water treated by the crystallization reaction unit, stabilize the effluent load of the intelligent control system, and adapt to the continuous operation of the process. The effluent fluoride ion meter 132 is used to collect the effluent fluoride ion concentration online in real time. Figure 1 The black arrow in the middle water outlet unit 13 indicates the direction of the water flow after crystallization and defluorination treatment.

[0062] The seed crystal regeneration and reuse unit 14 is used to classify and separate the calcium fluoride crystal mixture generated by the intelligent control method for defluorination of fluoride-containing wastewater crystallization provided in this application embodiment, and to regenerate the seed crystals that do not meet the requirements for crystal reuse. Its core components include a hydrocyclone separator 141, a seed crystal regeneration device 142, a seed crystal dosing pump 143, and a seed crystal dosing control valve 144. The hydrocyclone separator 141 is used to classify and separate the calcium fluoride crystal mixture discharged from the crystallization reaction unit, accurately separating large-particle crystals that meet the standards and small-particle crystals that require regeneration. The separated large-particle crystals are collected and directly recycled as crystal products, while the separated small-particle crystals are guided and conveyed to the seed crystal regeneration device 142, realizing the classification and utilization of the crystals. The seed crystal regeneration device 142 is used to receive the small-particle crystals conveyed by the hydrocyclone separator 141, and to regenerate the small-particle crystals (such as surface impurity cleaning, particle size regularization, and activity activation), restoring the crystallization activity of the small-particle crystals and restoring their crystallization ability to achieve crystal regeneration. The standard for seed reuse reduces the amount of fresh seed crystals added, lowering operating costs. The seed crystal addition pump 143 provides power to transport qualified regenerated seed crystals, quantitatively and continuously delivering the regenerated seed crystals to the crystallization reaction unit, realizing the recycling of seed crystals, ensuring stable seed crystal concentration within the crystallization reaction unit, and enhancing the nucleation and growth effect of calcium fluoride crystals. The seed crystal addition control valve 144 controls the opening and closing of the seed reuse pipeline, adjusting the seed addition opening and amount. In conjunction with the intelligent control system and the real-time operating conditions of the crystallization reaction unit, it precisely matches the seed addition requirements, ensuring uniform and stable seed addition, achieving refined control of seed reuse, and preventing pipeline blockage or excessive or insufficient seed addition from affecting the crystallization effect. Figure 1 The black arrow in the seed crystal regeneration and reuse unit 14 indicates the flow direction of the large crystal particles transported by the cyclone separator 141.

[0063] For example, one possible implementation of the seed crystal regeneration device 142 for regenerating small-particle crystals is as follows: the seed crystal regeneration device regenerates the small-particle crystals through chemical cleaning (such as soaking in NaOH solution) and hot air drying to remove surface impurities and restore crystal nucleus activity. The regenerated seed crystals are then reused in the crystallization reaction tank via a seed crystal addition pump to ensure that the crystal nucleus concentration and activity are maintained within the process requirements.

[0064] For example, one possible implementation of the hydrocyclone separator 141 for the graded separation of the calcium fluoride crystal mixture discharged from the crystallization reaction unit is as follows: crystals with a particle size greater than or equal to 50 μm are identified as large particles and can be directly used as products; crystals with a particle size greater than 10 μm and less than or equal to 40 μm are identified as small particles and are reused after regeneration treatment by a seed crystal regeneration device; crystals with a particle size less than 10 μm are collected and treated separately to avoid affecting the stability of the reaction system and further reduce sludge production and treatment costs.

[0065] The intelligent control unit 15 is used to execute the intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment. Its core components include a sensing-acquisition layer 151, a preprocessing layer 152, a model layer 153, an execution-feedback layer 154, and a main control station 155. The sensing-acquisition layer 151 is used to collect initial multi-source time-series data corresponding to the influent unit 10, the reagent dosing unit 11, the crystallization reaction unit 12, and the effluent unit 13, and transmit the collected initial multi-source time-series data to a local edge controller deployed within the preprocessing layer. The preprocessing layer 152 is used to perform preprocessing on the received initial multi-source time-series data through the local edge controller, including outlier removal, missing value completion, data normalization, and multi-parameter time-series alignment, outputting preprocessed standardized multi-source time-series data, and storing the standardized multi-source time-series data in a local edge database. This local edge database is communicatively connected to a cloud server, which can synchronously obtain operating data from the local edge database in real time. The model layer 153 is estimated using a digital twin. The system consists of a model, an LSTM prediction model, and a reinforcement learning decision model, used to output target control parameters based on standardized multi-source time-series data; the execution-feedback layer 154 is used to receive control decision commands corresponding to the target control parameters, and control equipment such as the reagent dosing control valve 112, the seed crystal dosing control valve 144, the stirring device 121, and the main microchannel control switch 122 according to the control decision commands, while collecting real-time system operation feedback data of the fluoride-containing wastewater crystallization defluorination process and transmitting it back to the sensing-acquisition layer 151 to form a closed-loop feedback; the main control station 155 is used to realize human-machine interface operation, real-time display of system operation status, equipment fault alarm prompts, model parameter configuration, and overall scheduling and control command issuance of the operation logic of each functional unit, supporting intelligent management and control of the entire crystallization defluorination process.

[0066] The following is based on Figure 1 The digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization, as shown in the diagram, serves as the main execution entity, or... Figure 1The intelligent control unit in the digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater crystallization, as shown in the illustration, serves as the execution entity. Specific implementation methods of the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application will be described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0067] Figure 2 A flowchart illustrating the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater via crystallization, as provided in this application embodiment. Figure 1 .like Figure 2 As shown, a specific implementation of this intelligent control method for defluoridation through crystallization in fluoride-containing wastewater may include the following steps:

[0068] S201, acquire standardized multi-source time-series data of the fluoride removal crystallization process of fluoride-containing wastewater operating within the current time window.

[0069] For example, standardized multi-source time-series data refers to various types of pre-processed time-series data collected from a fluoride-containing wastewater crystallization and defluorination intelligent control system. Standardized multi-source time-series data includes, but is not limited to, standardized influent fluoride ion concentration (mg / L), standardized influent flow rate (m³ / h), standardized influent temperature (°C), standardized calcium salt dosage (L / h), standardized calcium salt concentration (g / L), standardized pH value of the water within the crystallization reaction unit, standardized crystal particle size distribution (μm), standardized crystallization concentration (kg / m³), standardized stirring intensity (r / min), standardized number of microchannel reaction columns opened, and standardized effluent fluoride ion concentration (mg / L). The standardized crystal particle size distribution includes the particle size range distribution and average crystal particle size of the crystallization system.

[0070] For example, the current time window can be the time period corresponding to the steady-state operation of fluoride removal crystallization in fluoride-containing wastewater. This application does not limit the size of the time window; it can be determined based on actual application requirements.

[0071] For example, one possible implementation of this step is to obtain the above... Figure 1The parameters shown are: initial influent parameters collected by the influent unit within the current time window (e.g., initial influent fluoride ion concentration collected by the influent fluoride ion meter, initial influent flow rate collected by the influent flow meter, initial influent temperature collected by the influent temperature sensor); initial reagent dosing parameters collected by the reagent dosing unit within the current time window (e.g., initial calcium salt reagent dosage collected by the reagent flow meter, initial calcium salt reagent concentration collected by the reagent concentration detector); and initial crystallization reaction parameters collected by the crystallization reaction unit within the current time window (e.g., initial water pH value collected by the pH meter, initial crystallization particle size distribution collected by the crystallization particle size analyzer). The initial crystallization concentration (including seed concentration) collected by the crystallization concentration detector within the current time window, the initial stirring intensity collected by the stirring intensity sensor within the current time window, the initial number of microchannel reaction columns opened within the current time window, and the initial effluent parameters collected by the effluent unit within the current time window (such as the initial effluent fluoride concentration collected by the effluent fluoride meter) are processed sequentially. The initial influent parameters, initial reagent dosing parameters, initial crystallization reaction parameters, and initial effluent parameters are processed sequentially with outlier removal, missing value completion, data normalization, and multi-parameter time series alignment to remove invalid data, correct deviations, and obtain standardized multi-source time series data.

[0072] Among them, the local clocks of the online detection devices such as the influent fluoride ion meter, influent flow meter, influent temperature sensor, reagent flow meter, reagent concentration detector, pH meter, crystal particle size analyzer, crystal concentration detector, stirring intensity sensor and effluent fluoride ion meter are all calibrated based on a unified system master clock (such as Coordinated Universal Time, or UTC) clock, and all online detection devices synchronously collect data based on the data acquisition timestamp triggered by the system master clock.

[0073] For example, the initial influent parameters, initial reagent dosing parameters, initial crystallization reaction parameters, and initial effluent parameters collected by all the aforementioned online monitoring devices can be transmitted in real time to the local edge controller for storage. This ensures the continuity of time-series data and effectively avoids data loss or missing information. Furthermore, the standardized multi-source time-series data obtained after preprocessing can be transmitted in real time to the local edge time-series database for storage, enabling real-time on-site control. Simultaneously, the data stored in the local edge time-series database is periodically uploaded to the cloud server connected to the communication connection (e.g., the data upload frequency can be once per hour) to form a historical dataset, providing data support for subsequent model iteration training and fault mode library construction. The local edge controller and the local edge time-series database are both deployed at the on-site edge.

[0074] S202, standardized multi-source time series data are input into the digital twin estimation model and the LSTM prediction model respectively to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model.

[0075] The digital twin estimation model is a virtual model built based on a combination of physical mechanisms and data-driven approaches. It is used to simulate and estimate key kinetic parameters in the crystallization and defluorination reaction process in real time, namely the crystallization kinetic precipitation rate. The crystallization kinetic precipitation rate is a core parameter that determines the treatment effect of fluoride-containing wastewater and the quality of calcium fluoride crystals.

[0076] For example, one possible implementation involves constructing a digital twin estimation model of the entire crystallization defluorination process based on the crystallization defluorination reaction kinetic equation (which can also be described as the induced crystallization reaction kinetic equation) and standardized multi-source time-series data. This model enables real-time mapping of influent load, crystallization reaction parameters, crystallization state, and effluent quality. Specifically, constructing the digital twin estimation model of the entire crystallization defluorination process may include the following steps:

[0077] 1) Constructing a physical mechanism model: Based on the thermodynamic and kinetic characteristics of the core reaction principle of crystallization for defluorination, and combined with material conservation and energy conservation, a physical mechanism model is constructed.

[0078] The core reaction principle of crystallization for defluorination can be expressed as follows: .

[0079] 2) Model composition division: The digital twin estimation model consists of three parts: ion solubility thermodynamics model, crystallization kinetics model and time series data correction model. The three parts work together to ensure the estimation accuracy and adaptability of the digital twin estimation model to the operating conditions.

[0080] For example, the target crystallization kinetic precipitation rate is calculated by a digital twin estimation model based on the current system state, i.e., standardized multi-source time-series data. It is a key indicator used to characterize the operating efficiency and treatment effect of the crystallization defluorination process. This target crystallization kinetic precipitation rate is directly related to the formation rate of calcium fluoride crystals and the removal efficiency of fluoride ions, and is an important basis for optimizing the crystallization defluorination process.

[0081] Among them, the time series data correction model can adopt the LSTM correction model.

[0082] 3) Model parameter setting: The initial parameters of the model can be configured based on the actual working conditions on site, including the effective volume of the crystallization reactor, the activity of the crystal seed, and the reaction efficiency of the reagent. Subsequently, the model parameters are continuously corrected by the real-time system operation feedback data on site, so that the model maintains a high degree of matching with the actual operating conditions.

[0083] Among them, the effective volume of the crystallization reactor refers to the actual effective volume within the crystallization reaction unit that can accommodate the mixed system of fluoride-containing wastewater, calcium salt reagent, and seed crystals for reaction; seed crystal activity refers to the ability of calcium fluoride seed crystals to induce crystal nucleation and accelerate the growth of calcium fluoride crystals within the system, which directly determines the crystal nucleation rate and crystal growth efficiency; reagent reaction efficiency refers to the proportion of the added calcium salt reagent that actually participates in the calcium fluoride formation reaction and combines with fluoride ions to precipitate.

[0084] Among them, the LSTM prediction model is a prediction model built on the LSTM neural network. It is used to predict the key effluent indicators of the intelligent control system for defluoridation of fluoride-containing wastewater based on standardized multi-source time series data, that is, to predict the effluent indicators, and to provide input basis for the reinforcement learning decision model.

[0085] For example, the LSTM prediction model can adopt a 3-layer LSTM neural network structure; the input layer dimension is set to the data dimension corresponding to the standardized multi-source time series data, the number of hidden layer nodes is set to 64, the output layer dimension is set to the data dimension corresponding to the key water discharge indicators, the activation function is the Rectified Linear Unit (ReLU) function, the optimizer is the Adaptive Moment Estimation (Adam) optimizer, and the learning rate is set to 0.001.

[0086] For example, the predicted effluent indicators can be the effluent indicators of the next time window adjacent to the current time window, including but not limited to the predicted effluent fluoride concentration (unit: mg / L), predicted crystal particle size distribution (including average crystal particle size, unit: μm), calcium salt reagent consumption (unit: kg / h), etc.

[0087] One possible implementation of this step is to: input standardized multi-source time-series data into a digital twin estimation model to estimate the crystallization kinetics sedimentation rate, thereby obtaining the target crystallization kinetics sedimentation rate output by the digital twin estimation model; and input standardized multi-source time-series data into an LSTM prediction model to predict key effluent indicators, thereby obtaining the predicted effluent indicators output by the LSTM prediction model.

[0088] S203, the target crystallization kinetic precipitation rate and predicted effluent indicators are input into the reinforcement learning decision model to calculate the control parameters, and the target control parameters output by the reinforcement learning decision model are obtained.

[0089] Among them, the reinforcement learning decision model learns the optimal control strategy autonomously based on historical system operation data, and outputs the optimal control parameters, i.e. target control parameters, of the crystallization defluorination intelligent control system in combination with the system operation objectives, so as to achieve precise and intelligent control of the process.

[0090] For example, the reinforcement learning decision model can employ the Proximal Policy Optimization (PPO) algorithm, constructed based on the Offline Reinforcement Learning (Offline RL) paradigm. The input to the reinforcement learning decision model can be the real-time operating status of the system, water quality parameters, and process conditions, while the output is the control parameters.

[0091] For example, in one possible implementation: the target crystallization kinetic sedimentation rate is used as the core feedback parameter, and the predicted effluent indicators (effluent fluoride concentration, predicted crystal particle size distribution, and calcium salt reagent consumption) output by the LSTM prediction model are input into the reinforcement learning decision model. This allows the reinforcement learning decision model to output the target control parameters in real time, forming a closed-loop control logic of "collection-rate calculation-decision-control-feedback", ensuring that the crystallization kinetic sedimentation rate stably approaches and is maintained within the set optimal working range.

[0092] For example, the target control parameters include, but are not limited to: Ca / F molar ratio, pH setpoint, seed crystal dosage, stirring intensity, and the number of microchannel reaction columns opened. Specifically, the Ca / F molar ratio ranges from 0.7 to 0.9 (to ensure sufficient fluoride ion reaction while avoiding excessive calcium salt reagent), the pH setpoint ranges from 7.0 to 7.5 (within this range, calcium fluoride crystallization stability is optimal), the seed crystal dosage ranges from 50 mg / L to 100 mg / L (to maintain the number of crystal nuclei for crystal growth), the stirring intensity ranges from 150 r / min to 200 r / min (to ensure thorough mixing of calcium salt reagent and fluoride-containing wastewater while avoiding seed crystal breakage), and the number of microchannel reaction columns opened can be dynamically adjusted according to real-time operating conditions.

[0093] S204, based on the target control parameters, controls the crystallization and defluorination process of fluoride-containing wastewater in the next time window adjacent to the current time window.

[0094] In this step, one possible implementation method is to send the target control parameters to the field actuator (as described above). Figure 1 The main control station shown in the figure controls the opening degree of the reagent dosing control valve 112, the stirring intensity of the stirring device 121, the opening degree of the seed crystal dosing control valve 144, and the number of multiple microchannel reaction columns 123 opened through the actuators. By coordinating the above process equipment, the precipitation rate of calcium fluoride crystallization kinetics is controlled in real time, so that the precipitation rate of crystallization kinetics stabilizes and is maintained within the set optimal working range.

[0095] The number of microchannel reaction columns 123 that are activated can be controlled in the following way: The number and activation mode of the activated microchannel reaction columns are adjusted via the main microchannel control switch 122. The activation process employs an alternating, staggered activation method to prevent simultaneous activation and deactivation, thus avoiding sudden changes in operating conditions, maintaining the stability of the crystallization reaction system, and mitigating the problem of excessively high or low local reaction intensities. This control method can directly change the effective reaction volume and hydraulic residence time of the system, thereby regulating the solution supersaturation ratio and the precipitation rate of calcium fluoride crystallization kinetics. The specific regulation mechanism is as follows:

[0096] 1) When the number of microchannel reaction columns is increased, the reaction space and residence time for crystallization and fluoride removal increase, the contact and reaction between fluoride ions and calcium ions are more complete, the supersaturation ratio of the solution increases, and the crystallization kinetics and precipitation rate are accelerated.

[0097] 2) When the number of microchannel reaction columns opened decreases, the effective reaction volume and residence time decrease, the sufficiency of the crystallization and defluorination reaction decreases, the local supersaturation ratio decreases, and the crystallization kinetic precipitation rate slows down accordingly.

[0098] This application embodiment, by employing a control method that combines quantity adjustment with staggered operation of multiple microchannel reaction columns, can achieve rapid, precise, and stable regulation of the crystallization kinetic precipitation rate without significantly adjusting the calcium salt reagent dosage.

[0099] This application embodiment employs a multi-model collaborative architecture, acquires standardized multi-source time-series data, combines the target crystallization kinetic sedimentation rate output by the digital twin estimation model with the predicted effluent indicators output by the LSTM prediction model, and utilizes a reinforcement learning decision model to calculate target control parameters. This enables dynamic control of the kinetic characteristics of the crystallization and defluoridation process in fluoride-containing wastewater, improves the control accuracy of the system's crystallization kinetic sedimentation rate, and effectively overcomes the shortcomings of traditional manual experience-based fixed control parameter setting leading to kinetic inaccuracies and low control accuracy of the system's crystallization kinetic sedimentation rate. It can ensure that the effluent fluoride concentration stably meets the standards, while reducing the consumption of calcium salt reagents and the amount of sludge generated, significantly improving the system's operational stability, reliability, and economy.

[0100] Optionally, the digital twin estimation model includes an ion solubility thermodynamic model, a crystallization kinetic model, and a time-series data correction model, with standardized multi-source time-series data including crystallization reaction temperature.

[0101] The crystallization reaction temperature can be determined by a temperature sensor installed in the crystallization reaction unit. Figure 1 The initial crystallization reaction temperature (not shown) is obtained after preprocessing the data collected in real time; alternatively, it can be obtained after preprocessing the initial inlet water temperature collected in real time by the inlet water temperature sensor installed in the inlet water unit.

[0102] It is understandable that since the crystallization and defluorination of fluoride-containing wastewater usually operates under normal temperature conditions, the temperature difference between the influent and the crystallization reaction tank is small. Therefore, the influent temperature can be used to approximate the crystallization reaction temperature. The effect of this approximation method on the correction result of the ion solubility product is negligible.

[0103] The following is combined with Figure 3 A detailed explanation is provided on a specific implementation method for inputting standardized multi-source time-series data into the digital twin estimation model in step S202 to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model.

[0104] Figure 3 A flowchart illustrating the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater via crystallization, as provided in this application embodiment. Figure 2 .like Figure 3 As shown, in this intelligent control method for defluoridation of fluoride-containing wastewater through crystallization, a specific implementation method for inputting standardized multi-source time-series data into a digital twin estimation model to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model may include the following steps:

[0105] S301, based on the thermodynamic model of ion solubility product, calculates the calcium fluoride solubility product corresponding to the crystallization reaction temperature according to the crystallization reaction temperature.

[0106] For example, in one possible implementation, the solubility product of calcium fluoride at the crystallization reaction temperature is calculated based on the solubility product of calcium fluoride at 25°C.

[0107] For example, the calcium fluoride solubility product corresponding to the crystallization reaction temperature can be expressed by the following formula:

[0108]

[0109] in, This represents the calcium fluoride solubility product corresponding to the crystallization reaction temperature, in units of... ; This represents the solubility product of calcium fluoride at 25℃, and its value is... Converted to The corresponding value is ; This represents the enthalpy change of the calcium fluoride crystallization defluorination reaction, and is taken as a constant (the specific value is adjusted according to the type of reagent used on site, such as when CaCl2 is used as the reagent). Values R represents the gas constant, which takes the value of ); T represents the real-time thermodynamic temperature (unit: K) within the crystallization reaction tank, which is obtained by converting the crystallization reaction temperature T1 (unit: °C). .

[0110] S302, based on a crystallization kinetics model, calculates the simulated crystallization kinetics precipitation rate using standardized multi-source time-series data and the calcium fluoride solubility product.

[0111] Optionally, the standardized multi-source time-series data also includes influent parameters of fluoride-containing wastewater, calcium salt dosage parameters, crystallization reaction process parameters, and current crystallization kinetics precipitation rate. Among these, influent parameters include standardized influent fluoride ion concentration (mg / L) and standardized influent flow rate (m³ / h); calcium salt dosage parameters include standardized calcium salt dosage (L / h) and standardized calcium salt concentration (g / L); and crystallization reaction process parameters include standardized crystal particle size distribution (μm) and standardized crystallization concentration (kg / m³).

[0112] Optionally, this step may include the following steps:

[0113] S3021, based on the crystallization kinetics model, calculates the current calcium ion concentration and current fluoride ion concentration in the crystallization reaction tank according to the influent parameters of fluoride-containing wastewater, the calcium salt reagent addition parameters, and the current crystallization kinetics precipitation rate.

[0114] For example, based on the crystallization kinetics model, the current calcium ion concentration is calculated according to the standardized calcium salt dosage (unit: L / h), standardized calcium salt concentration (unit: g / L), current crystallization kinetic precipitation rate (unit: kg / s), and standardized influent flow rate (unit: m³ / h).

[0115] For example, the current calcium ion concentration can be expressed by the following formula:

[0116]

[0117] in, This represents the current calcium ion concentration at time t (unit: mg / L). This indicates the standardized calcium salt dosage (unit: L / h). This indicates the standardized concentration of calcium salt reagents (unit: g / L). This represents the current crystallization kinetic precipitation rate (unit: kg / s) at the current time t. This indicates the standardized influent flow rate (unit: m³ / h).

[0118] For example, based on a crystallization kinetics model, the current fluoride ion concentration is calculated according to the standardized influent fluoride ion concentration (unit: mg / L) and the standardized influent flow rate (unit: m³ / h).

[0119] For example, the current fluoride ion concentration can be expressed by the following formula:

[0120]

[0121] in, This represents the current fluoride ion concentration at time t (unit: mg / L). The standard influent fluoride concentration is expressed in mg / L, and the coefficient 2 indicates that 1 mol of calcium fluoride precipitate corresponds to the consumption of 2 mol of fluoride ions.

[0122] For example, in the embodiments of this application, the initial calcium ion concentration can be directly selected from the standardized calcium salt reagent concentration, and the initial fluoride ion concentration can be directly selected from the standardized influent fluoride ion concentration, as the initial calculation conditions for the crystallization kinetic model.

[0123] S3022, calculate the supersaturation ratio based on the current calcium ion concentration, the current fluoride ion concentration, and the calcium fluoride solubility product.

[0124] One possible implementation of this step is to calculate the ion activity product based on the current calcium ion concentration and the current fluoride ion concentration, and further calculate the supersaturation ratio based on the ion activity product and the calcium fluoride solubility product.

[0125] For example, the ion activity product can be expressed by the following formula:

[0126]

[0127] in, This represents the ion activity product at the current time t.

[0128] For example, the supersaturation ratio can be expressed by the following formula:

[0129]

[0130] Where S(t) represents the oversaturation ratio at the current time t.

[0131] It is understood that the embodiments of this application use relative supersaturation to define the supersaturation ratio, while in engineering, ion concentration is directly used to approximate activity in order to ensure the convenience and practicality of calculation.

[0132] S3023, calculates the simulated crystallization kinetic precipitation rate based on the supersaturation ratio, crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetic model.

[0133] For example, the crystallization reaction process parameters include the normalized crystallite size distribution (unit: μm) and the normalized crystallization concentration (unit: kg / m³).

[0134] In one possible implementation, for example, a crystal growth kinetics formula is used, combined with a standardized crystal grain size distribution and a standardized crystal concentration, to calculate the simulated crystal growth kinetics precipitation rate.

[0135] For example, the precipitation rate in simulated crystallization kinetics can be expressed by the following formula:

[0136]

[0137] in, This represents the precipitation rate of the simulated crystallization kinetics at the current time t. This represents the current growth rate compensation constant at time t (unit: kg / m³). 2 / s), whose value can be set according to the crystallization type on site (e.g., crystallization type is During crystallization, Pick ); g represents the growth stage, and its value can be adjusted according to the on-site working conditions. Generally, the value range can be 1.5 to 2.0. In the embodiment of this application, the value can be 1.8. This represents the total specific surface area of ​​the crystals at the current time t (unit: m²).

[0138] For example, the total specific surface area of ​​the crystals is calculated based on the normalized crystal grain size distribution and the normalized crystal concentration.

[0139] For example, the total specific surface area of ​​crystals can be expressed by the following formula:

[0140]

[0141] in, Indicates the standardized crystallization concentration (unit: kg / m³). This represents the effective volume of the crystallization reactor (unit: m³), ​​and its value is set according to the size of the crystallization reaction unit on site; it is a fixed value. This indicates the density of calcium fluoride, with a value of 3180 kg / m³. This indicates the average crystal grain size (unit: μm).

[0142] It is understood that in the crystallization kinetics model provided in this application, the supersaturation ratio is calculated using the current calcium ion concentration, the current fluoride ion concentration, and the calcium fluoride solubility product. Combined with the growth rate compensation constant, the simulated crystallization kinetics precipitation rate is iteratively calculated. This calculation logic quantifies the relationship between the supersaturation state of the reaction system and the crystal growth rate, making the simulation results more closely resemble actual working conditions and providing precise dynamic feedback for subsequent regulation.

[0143] S303, based on the time-series data correction model, corrects the simulated crystallization kinetics precipitation rate to obtain the target crystallization kinetics precipitation rate.

[0144] For example, one possible implementation is to use a time-series data correction model, introducing an LSTM time-series correction term during the calculation of the crystallization kinetics precipitation rate to correct the calculated simulated crystallization kinetics precipitation rate, and finally outputting the corrected target crystallization kinetics precipitation rate, ensuring that the calculation error of the target crystallization kinetics precipitation rate is controlled within a certain range. Within this range, reliable support is provided for precise control of subsequent processes.

[0145] For example, the target crystallization kinetic precipitation rate can be expressed by the following formula:

[0146]

[0147] in, Indicates the target crystallization kinetics precipitation rate. This represents an LSTM timing correction network. This represents historical system operation data (such as historical pH value of water in the crystallization reaction unit, historical influent flow rate, historical crystal particle size distribution, etc.).

[0148] In this embodiment, based on an ion solubility product thermodynamic model, the calcium fluoride solubility product corresponding to the crystallization reaction temperature is calculated. Based on a crystallization kinetic model, the simulated crystallization kinetic precipitation rate is calculated using standardized multi-source time-series data and the calcium fluoride solubility product. The simulated crystallization kinetic precipitation rate is then corrected using a time-series data correction model to obtain the target crystallization kinetic precipitation rate. This embodiment employs a hierarchical, progressive, and refined modeling and correction mechanism, enabling the digital twin estimation model to output more accurate and reliable process state characterization information (i.e., the target crystallization kinetic precipitation rate). This provides high-quality input features for the subsequent reinforcement learning decision model, thereby significantly improving the system's decision-making accuracy and control stability.

[0149] It is understood that, in this embodiment, the digital twin estimation model consists of an ion solubility product thermodynamic model, a crystallization kinetics model, and a time-series data correction model. The ion solubility product thermodynamic model calculates the calcium fluoride solubility product based on the crystallization reaction temperature. The crystallization kinetics model calculates the simulated crystallization kinetics precipitation rate by combining influent parameters, reagent dosing parameters, and the current crystallization kinetics precipitation rate. The time-series data correction model corrects the simulated crystallization kinetics precipitation rate by incorporating historical operating data, ultimately outputting a highly accurate target crystallization kinetics precipitation rate. This synergistic mechanism, through the combination of physical mechanisms and data-driven approaches, significantly improves the accuracy of crystallization kinetics precipitation rate estimation, providing a reliable basis for subsequent regulation.

[0150] Optionally, to further ensure the high degree of matching between the target crystallization kinetic sedimentation rate output by the digital twin estimation model and the actual operating conditions, and to improve the control accuracy, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: calculating the crystallization kinetic sedimentation rate deviation between the current crystallization kinetic sedimentation rate and the simulated crystallization kinetic sedimentation rate; using a PID control algorithm, calculating the compensation adjustment amount of the current growth rate compensation constant based on the crystallization kinetic sedimentation rate deviation; and accumulating the compensation adjustment amount to the current growth rate compensation constant to obtain the target growth rate compensation constant.

[0151] For example, the compensation adjustment amount of the current growth rate compensation constant can be expressed by the following formula:

[0152]

[0153] in, This represents the compensation adjustment amount corresponding to the current time t (unit: kg / m). 2 / s); This represents the proportionality coefficient (unitless), with a value range of 0.3-0.4 and a baseline value of 0.4. This represents the integral coefficient (unitless), with a value range of 0.05-0.1 and a baseline value of 0.08. This represents the differential coefficient (unitless), with a value range of 0.1-0.2 and a baseline value of 0.15. This represents the deviation of the crystallization kinetic precipitation rate (unit: kg / s) between the current crystallization kinetic precipitation rate and the simulated crystallization kinetic precipitation rate at the current time t.

[0154] For example, the proportional coefficient, integral coefficient, and differential coefficient can be set according to the system response characteristics (such as the rate of change of crystallization rate and the cumulative trend of deviation) to ensure that the model parameters are corrected in real time and improve the long-term stability of the kinetic estimation.

[0155] For example, the deviation of the crystallization kinetics precipitation rate can be expressed by the following formula:

[0156]

[0157] For example, the target growth rate compensation constant can be expressed by the following formula:

[0158]

[0159] in, This represents the target growth rate compensation constant corresponding to the current time t.

[0160] In this embodiment, the deviation between the current crystallization kinetic precipitation rate and the simulated crystallization kinetic precipitation rate is calculated. A PID control algorithm is then used to calculate the compensation adjustment amount of the current growth rate compensation constant based on this deviation. This compensation adjustment amount is further accumulated into the current growth rate compensation constant to obtain the target growth rate compensation constant. This target growth rate compensation constant is fed back to the crystallization kinetic model for subsequent calculations of the simulated crystallization kinetic precipitation rate. Through this closed-loop feedback mechanism, and the synergistic effect of proportional, integral, and derivative operations, the key parameters within the digital twin estimation model can be adaptively adjusted in real time according to the actual operating conditions. This ensures that the simulation results are highly consistent with the actual operating state, improving the long-term stability and prediction accuracy of the model.

[0161] Optionally, step S3023, which calculates the simulated crystallization kinetics precipitation rate based on the supersaturation ratio, crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetics model, can be implemented by iteratively calculating the simulated crystallization kinetics precipitation rate based on the supersaturation ratio, crystallization reaction process parameters, and the target growth rate compensation constant. The specific implementation of this embodiment is similar to that described above and will not be repeated here.

[0162] Optionally, step S203, which inputs the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model to calculate the control parameters and obtain the target control parameters output by the reinforcement learning decision model, can be implemented by: inputting the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model, taking the target crystallization kinetic sedimentation rate being maintained within the preset crystallization kinetic sedimentation rate range as the core control objective, and combining the compliance constraints of the predicted effluent index to output the target control parameters.

[0163] For example, the preset crystallization kinetics precipitation rate range can be The preset crystallization kinetic precipitation rate range can balance fluoride ion removal efficiency, crystal quality, and calcium salt reagent consumption, and is the core objective of the entire closed-loop control.

[0164] For example, the compliance constraints for predicted effluent indicators refer to the fact that while optimizing the core control objective of maintaining the target crystallization kinetic sedimentation rate within a preset crystallization kinetic sedimentation rate range, the reinforcement learning decision model must also ensure that its decisions do not cause the predicted effluent indicators to exceed the preset compliance range. Compliance constraints may include effluent fluoride ion concentration, water pH value, predicted crystal particle size distribution, and calcium salt reagent consumption, etc.

[0165] For example, the target control parameter can be a continuous value or a discrete value. This application does not limit this, and it can be determined according to the actual application requirements.

[0166] It is understood that, in this embodiment, by inputting the target crystallization kinetic sedimentation rate and the predicted effluent indicators into the reinforcement learning decision model, and taking the maintenance of the target crystallization kinetic sedimentation rate within a preset range as the core control objective, the model outputs target control parameters by combining the compliance constraints of the predicted effluent indicators, thereby improving the system's control accuracy. Simultaneously, the multi-weighted reward function balances effluent indicator compliance, reagent consumption, and system stability, ensuring that the control parameters optimize operational economy while meeting environmental requirements, thus avoiding suboptimal solutions caused by single-objective optimization.

[0167] Optionally, the target control parameters include the Ca / F molar ratio, pH setpoint, seed crystal dosage, stirring intensity, and the number of microchannel reaction columns opened. The specific values ​​for the Ca / F molar ratio, pH setpoint, seed crystal dosage, and stirring intensity are similar to those described above and will not be repeated here.

[0168] It should be noted that, in order to achieve precise control, different subdivisions of the precipitation rate of the target crystallization kinetics can be defined to formulate a microchannel reaction column opening quantity control strategy and a seed crystal addition quantity control strategy to ensure a stable and efficient crystallization process.

[0169] Optionally, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: determining the number of microchannel reaction columns to be opened and the spatial distribution pattern of the microchannel reaction columns according to the target crystallization kinetic precipitation rate, wherein the number of microchannel reaction columns to be opened is determined based on the total number of microchannel reaction columns, and the opened microchannel reaction columns are staggered at intervals.

[0170] Understandably, the number of microchannel reaction columns that are opened is primarily used to regulate the effective reaction volume and hydraulic residence time of the system, directly affecting the supersaturation within the crystallization reaction system, and thus adjusting the kinetic crystallization precipitation rate. The principle of intermittent and staggered operation must be strictly adopted to prevent centralized operation.

[0171] For example, one possible implementation of determining the number of microchannel reaction columns to be opened and the spatial distribution pattern of the microchannel reaction columns based on the target crystallization kinetic precipitation rate is as follows:

[0172] 1) When the target crystallization kinetic precipitation rate is less than or equal to When the target crystallization kinetic sedimentation rate is severely low: significantly increase the number of microchannel reaction columns that are opened, with the number of opened microchannel reaction columns being 80% to 100% of the total number of microchannel reaction columns. The newly opened microchannel reaction columns should still be opened intermittently and staggered with each other, and evenly distributed with the original microchannel reaction columns that have been put into operation, so as to maximize the effective reaction volume of the system, extend the hydraulic residence time, improve the sufficiency of crystallization reaction, and rapidly increase the crystallization kinetic sedimentation rate.

[0173] 2) When the target crystallization kinetic precipitation rate is greater than and less than When the target crystallization kinetic precipitation rate is low and close to the lower limit of the optimal range: moderately increase the number of microchannel reaction columns to be opened, with the number of opened columns being 65% to 75% of the total number of microchannel reaction columns. The newly opened microchannel reaction columns should still be opened in an intermittent and staggered manner, and evenly distributed with the existing microchannel reaction columns to slightly increase the effective reaction volume of the system, slowly increase the crystallization kinetic precipitation rate, and avoid sudden changes in the crystallization kinetic precipitation rate.

[0174] 3) When the target crystallization kinetic precipitation rate is greater than or equal to and less than When the target crystallization kinetic precipitation rate is within the optimal range: keep the number of microchannel reaction columns open unchanged, and keep the number of open columns stable at 50%~60% of the total number of microchannel reaction columns. Continue to maintain the operation mode of intermittent opening and mutual staggering to ensure the stability of the effective reaction volume of the system and keep the kinetic crystallization precipitation rate within the optimal range for a long time.

[0175] 4) When the target crystallization kinetic precipitation rate is greater than or equal to and less than or equal to When the target crystallization kinetic precipitation rate is too high and close to the upper limit of the optimal range: moderately reduce the number of microchannel reaction columns to be opened, with the number of open columns being 35% to 45% of the total number of microchannel reaction columns. When reducing, prioritize closing continuously distributed microchannel reaction columns, retain intermittently opened microchannel reaction columns, slightly reduce the effective reaction volume of the system, and slowly reduce the crystallization kinetic precipitation rate.

[0176] 5) When the target crystallization kinetic precipitation rate is greater than or equal to When the target crystallization kinetic precipitation rate is significantly high: reduce the number of microchannel reaction columns that are turned on to 20% to 30% of the total number of microchannel reaction columns. Prioritize turning off densely distributed microchannel reaction columns to ensure that the remaining operational microchannel reaction columns are still staggered. This will minimize the effective reaction volume of the system, shorten the hydraulic residence time, and quickly suppress the crystallization reaction rate.

[0177] It should be noted that the opening ratio of the microchannel reaction columns mentioned above is based on the total number of microchannel reaction columns, specifying the exact proportion of the number of microchannel reaction columns in operation relative to the total number of microchannel reaction columns. On-site, the opening ratio range of the microchannel reaction columns can be fine-tuned according to the actual reactor specifications (such as the total number of microchannel reaction columns and the single column processing capacity) (the fine-tuning range shall not exceed ±5%). The commissioning operation of all microchannel reaction columns strictly follows the principle of intermittent opening and mutual staggering to ensure that the microchannel reaction columns in operation are evenly distributed regardless of the opening ratio, avoiding problems such as uneven reaction system and abnormal local supersaturation caused by concentrated opening or concentrated closing, and ensuring that the opening ratio of the microchannel reaction columns matches the reaction conditions.

[0178] Optionally, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: determining the target opening degree of the seed addition control valve and the corresponding seed addition amount based on the target crystallization kinetic precipitation rate, wherein the opening degree of the seed addition control valve is positively correlated with the seed addition amount.

[0179] Understandably, the seed crystal dosage is primarily used to regulate the number of crystal nuclei, directly affecting the crystallization concentration and thus controlling the crystallization kinetics and precipitation rate. By precisely controlling the opening of the seed crystal dosing control valve, the seed crystal dosage can be dynamically adjusted. The valve opening is positively correlated with the seed crystal dosage (i.e., the larger the valve opening, the larger the seed crystal dosage).

[0180] For example, one possible way to determine the target opening degree of the seed crystal addition control valve and the corresponding seed crystal addition amount based on the target crystallization kinetic precipitation rate is as follows:

[0181] 1) When the target crystallization kinetic precipitation rate is less than or equal to When the target crystallization kinetic precipitation rate is significantly low: Increase the seed crystal dosage substantially, controlling it at 90 mg / L - 100 mg / L, corresponding to adjusting the target opening of the seed crystal dosage control valve to 85%-95%. By increasing the valve opening, the seed crystal addition rate is increased, the number of crystal nuclei is increased, the driving force for crystal growth is strengthened, and the crystallization kinetic precipitation rate is rapidly increased, pushing the crystallization kinetic precipitation rate back to the optimal range. When adjusting the valve opening, a gradual adjustment mode should be adopted, with each adjustment not exceeding 5%, to avoid sudden changes in seed crystal dosage that could disrupt the crystallization system.

[0182] 2) When the target crystallization kinetic precipitation rate is greater than and less than When the target crystallization kinetic precipitation rate is low, close to the lower limit of the optimal range: moderately increase the seed crystal dosage, controlling it at 80 mg / L - 90 mg / L, corresponding to adjusting the target opening of the seed crystal dosage control valve to 75%-85%. By slightly increasing the valve opening of the seed crystal dosage control valve, the seed crystal addition rate is increased, the number of crystal nuclei is increased, and the driving force for crystal growth is slowly promoted, so that the crystallization kinetic precipitation rate steadily rises to the optimal range;

[0183] 3) When the target crystallization kinetic precipitation rate is greater than or equal to and less than When the target crystallization kinetic precipitation rate is within the optimal range: maintain the seed crystal dosage within a reasonable range of 60 mg / L-80 mg / L, corresponding to a stable target opening of the seed crystal dosage control valve at 60%-75%. No significant adjustments are needed; only minor adjustments to the valve opening (no more than 3%) based on the crystal particle size feedback, and simultaneously minor adjustments to the seed crystal dosage (no more than 5 mg / L). This ensures a stable number of crystal nuclei, maintains balanced crystal growth, and guarantees the average crystal particle size. 50μm;

[0184] 4) When the target crystallization kinetic precipitation rate is greater than or equal to and less than or equal to When the target crystallization kinetic precipitation rate is too high, approaching the upper limit of the optimal range: moderately reduce the seed crystal dosage, controlling it at 50 mg / L - 60 mg / L. Correspondingly, adjust the target opening of the seed crystal dosage control valve to 50%-60%. Slightly reduce the valve opening to decrease the number of crystal nuclei, gently inhibit the crystal growth rate, avoid explosive nucleation, and allow the crystallization kinetic precipitation rate to gradually return to the optimal range. When adjusting the valve opening, avoid making excessive adjustments at once to prevent a sudden decrease in seed crystal dosage from affecting crystallization stability.

[0185] 5) When the target crystallization kinetic precipitation rate is greater than or equal to When the target crystallization kinetic precipitation rate is significantly too high: maintain the seed crystal dosage at 50 mg / L (lower limit), and keep the target opening of the seed crystal dosage control valve stable at 40%-50% (minimum effective opening to ensure continuous addition of a small amount of seed crystals and avoid crystal nucleus fragmentation). By minimizing the valve opening, control the number of crystal nuclei, effectively suppress the explosive nucleation phenomenon, reduce the formation of fine crystals, rapidly reduce the crystallization kinetic precipitation rate, and avoid deterioration of crystal quality.

[0186] It should be noted that the seed crystal addition control valve is an intelligent regulating valve, and the valve opening degree can be continuously adjusted from 0% to 100%, with the valve opening degree accuracy controlled within... 1% ensures precise control of seed crystal dosage; the correspondence between valve opening and seed crystal dosage can be periodically calibrated according to actual working conditions such as on-site seed crystal concentration and pipeline pressure (e.g., calibrated every two weeks) to ensure the matching of valve opening and seed crystal dosage and avoid deviation in seed crystal dosage due to pipeline pressure fluctuations.

[0187] It is understood that in this embodiment, the number of microchannel reaction columns opened is dynamically adjusted based on the target crystallization kinetics precipitation rate, and an intermittent distribution pattern (such as maintaining a certain distance between adjacent reaction columns) is adopted to avoid abnormal local reaction intensity. The seed crystal dosage is adjusted by controlling the valve opening, and the valve opening is positively correlated with the seed crystal dosage. This control method improves the uniformity of the reaction system by optimizing the reaction space distribution and seed crystal concentration, reducing local supersaturation or nucleation anomalies, and significantly improving crystallization quality and processing efficiency.

[0188] Optionally, the LSTM prediction model provided in this application embodiment is trained in the following manner: acquiring first multi-source time series training data, which is a preprocessed historical standardized multi-source time series dataset; inputting the first multi-source time series training data into the LSTM prediction model to be trained to predict water discharge indicators, and obtaining the predicted values ​​of water discharge indicators output by the LSTM prediction model to be trained; using the mean square error between the predicted values ​​of water discharge indicators and the true values ​​of water discharge indicators in the first multi-source time series training data as the loss function, iteratively optimizing the model parameters of the LSTM prediction model to be trained until the loss function converges, and obtaining the LSTM prediction model.

[0189] For example, the first multi-source time series training data can be a historical normalized multi-source time series dataset stored on a cloud server. That is, the cloud server stores historical normalized multi-source time series data that has undergone outlier removal, missing value completion, data normalization, and multi-parameter time series alignment. This dataset serves as the training input for the LSTM prediction model and the reinforcement learning decision model, ensuring that model parameter optimization is based on high-quality data, thereby improving prediction and decision accuracy.

[0190] For example, in this embodiment, one possible implementation is as follows: the first multi-source time-series training data is divided into a training set, a validation set, and a test set in a ratio of 7:2:1; the training set data is input into the LSTM prediction model to be trained to predict the water discharge index, and the predicted value of the water discharge index output by the LSTM prediction model to be trained is obtained; the mean square error between the predicted value of the water discharge index and the true value of the water discharge index in the training set data is used as the loss function, and the backpropagation algorithm is used to iteratively optimize the model parameters until the loss function converges; the model is further validated and tested through the validation set and the test set in turn, and finally the LSTM prediction model is obtained.

[0191] For example, when training an LSTM prediction model, the number of iterations can be 100 to avoid overfitting.

[0192] In this embodiment, by acquiring first multi-source time-series training data, the first multi-source time-series training data is input into the LSTM prediction model to be trained to predict water discharge indicators, and the predicted values ​​of water discharge indicators output by the LSTM prediction model to be trained are obtained. The mean square error between the predicted values ​​of water discharge indicators and the true values ​​of water discharge indicators in the first multi-source time-series training data is used as the loss function to iteratively optimize the model parameters of the LSTM prediction model to be trained until the loss function converges, thereby obtaining the LSTM prediction model, improving the model accuracy, and providing high-precision and high-reliability prediction of water discharge indicators for subsequent reinforcement learning decision models, thereby significantly improving decision accuracy and control effect, and ensuring that the system can still operate stably and meet the water discharge requirements in complex and ever-changing environments.

[0193] Optionally, the reinforcement learning decision model provided in this application embodiment is trained in the following way: acquiring second multi-source time-series training data; inputting the second multi-source time-series training data into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model to be trained; taking the multi-weight reward function as the optimization objective, iteratively optimizing the model parameters of the reinforcement learning decision model to be trained according to the control parameters until the control parameters output by the reinforcement learning decision model to be trained can make the score of the multi-weight reward function stable within a preset score range, thereby obtaining the reinforcement learning decision model.

[0194] For example, the second multi-source time-series training data can be a historical, normalized multi-source time-series dataset stored on a cloud server. The second multi-source time-series training data can be the same as or different from the first multi-source time-series training data.

[0195] For example, the preset score range can be greater than 85 points.

[0196] For example, in this embodiment, one possible implementation is as follows: the second multi-source time-series training data is divided into a training set, a validation set, and a test set in a 7:2:1 ratio; using the PPO offline reinforcement learning paradigm, the training set data is input into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model; with the multi-weighted reward function as the optimization objective, the model parameters of the reinforcement learning decision model to be trained are iteratively optimized according to the control parameters until the control parameters output by the reinforcement learning decision model to be trained can make the score of the multi-weighted reward function stable within a preset score range; the model is further validated and tested sequentially through the validation set and the test set to finally obtain the reinforcement learning decision model.

[0197] Specifically, in addition to ensuring that the output adjustment parameters of the reinforcement learning decision model to be trained can stabilize the score of the multi-weighted reward function within a preset score range, the score fluctuation of the multi-weighted reward function must also be less than or equal to a preset score fluctuation threshold within a preset number of consecutive iterations. For example, the preset number of iterations can be 10, and the preset score fluctuation threshold can be 5 points. This application does not limit the specific values ​​of the preset number of iterations and the preset score fluctuation threshold; they can be determined according to actual application requirements.

[0198] It is understood that the embodiments of this application construct a dual AI model architecture of "LSTM prediction model + reinforcement learning decision model", complete model training through historical standardized multi-source time series datasets, realize accurate prediction of key water discharge indicators and intelligent output of optimal control parameters, which can replace the traditional manual experience parameter adjustment and fixed PID control method, thereby effectively improving the control accuracy and operation stability of the system.

[0199] Optionally, the LSTM prediction model and reinforcement learning decision model provided in this application embodiment can also be iteratively optimized in real time so that the LSTM prediction model, reinforcement learning decision model and digital twin estimation model can dynamically adapt to long-term fluctuations in real-time field operating conditions (such as changes in water quality, changes in calcium salt concentration, and decay of seed activity), thereby continuously improving the control accuracy and system operation stability.

[0200] One possible implementation involves using newly added historical standardized multi-source time-series data stored on a cloud server every 24 hours to perform online iterative optimization training on both the LSTM prediction model and the reinforcement learning decision model, thereby further improving model adaptation accuracy. Specifically, newly added historical standardized multi-source time-series data from the last 24 hours is obtained from the cloud server and added to the original training dataset. Using the same model training method as described above, online iterative optimization training is performed on both the LSTM prediction model and the reinforcement learning decision model to correct model parameters and improve the model's adaptability to real-time operating conditions. After training, the model performance is further validated and tested based on validation and test sets, ultimately yielding the iteratively optimized LSTM prediction model and reinforcement learning decision model. Specifically, when the prediction accuracy of the LSTM prediction model is greater than or equal to 95%, iterative training is terminated. If the prediction accuracy of the LSTM prediction model is less than 95%, iterative training continues until the performance requirements are met. When the multi-weighted reward function score of the reinforcement learning decision model is stable above 85, iterative training is terminated. If the multi-weighted reward function score of the reinforcement learning decision model does not meet the requirement of being stable above 85, iterative training continues until the performance requirements are met.

[0201] Optionally, the multi-weighted reward function includes weights for effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability; the total score for the weights for effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability is 100 points.

[0202] Understandably, the design of the multi-weighted reward function is intended to achieve multi-objective optimization, including meeting effluent standards, minimizing calcium salt reagent consumption, optimizing crystallization quality, and ensuring stable system operation, with an overall score of 100 points.

[0203] For example, in one possible implementation, the score allocation for the weights of effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability is as follows:

[0204] 1) Weighting of effluent indicators (40 points): When the effluent fluoride concentration is less than or equal to 8 mg / L, the weighting of effluent indicators is 40 points; for every 0.1 mg / L that exceeds the standard, 5 points are deducted from the weighting of effluent indicators; when the effluent fluoride concentration exceeds the standard by more than 1 mg / L, that is, when the effluent fluoride concentration is greater than or equal to 9 mg / L, the weighting of effluent indicators is 0 points.

[0205] 2) Calcium salt drug dosage optimization weight (30 points): When the consumption of calcium salt drugs is 10% or more lower than the historical average, the calcium salt drug dosage optimization weight is 30 points; when the consumption of calcium salt drugs is within ±10% of the historical average, the calcium salt drug dosage optimization weight score is calculated linearly as 15-25 points; when the consumption of calcium salt drugs is 10% or more higher than the historical average, 3 points are deducted for every 5% increase, until all points are deducted.

[0206] 3) Crystallization particle size weight (20 points): When the average crystallization particle size is greater than or equal to 50 μm (excellent crystallization quality, easy to settle), the crystallization particle size weight is 20 points; for every 5 μm decrease in the average crystallization particle size, the crystallization particle size weight is deducted 2 points; when the average crystallization particle size is less than or equal to 30 μm (deterioration of crystallization, increase of fine mud), the crystallization particle size weight is 0 points.

[0207] 4) System stability weight (10 points): Within every 30 minutes, when the pH value of the water in the crystallization reaction unit fluctuates less than or equal to 0.5 and the stirring intensity fluctuates less than or equal to 10 r / min, the system stability weight will be 10 points. When the pH value of the water exceeds the allowable fluctuation range by 0.1 or the stirring intensity fluctuates more than the allowable fluctuation range by 1 r / min, the system stability weight will be deducted by 1 point, until the score is deducted.

[0208] By configuring the score allocation rules of the multi-weighted reward function, it is possible to ensure that the output control parameters meet environmental protection standards and achieve energy-saving and consumption-reducing operation of calcium salt reagents.

[0209] Optionally, one possible implementation of step S201, obtaining the standardized multi-source time series data at the current moment, is as follows: obtaining the initial multi-source time series data at the current moment; preprocessing the initial multi-source time series data to obtain standardized multi-source time series data, wherein the preprocessing includes outlier removal, missing value completion, data normalization, and multi-parameter time series alignment.

[0210] The method for obtaining the initial multi-source time series data is similar to that described above, and will not be repeated here.

[0211] For example, one possible implementation of preprocessing initial multi-source time series data to obtain standardized multi-source time series data may include the following steps:

[0212] 1) Outlier removal processing of initial multi-source time series data: using a method based on... The principle is combined with the isolated forest algorithm to remove outlier data caused by detection equipment failure, field interference (such as voltage fluctuations, chemical impacts), etc., resulting in multi-source time-series data after outlier removal. The principle is used to identify extreme values ​​that deviate from the normal range of data (such as values ​​exceeding the mean). (Data with a standard deviation of 10 ...

[0213] 2) Missing value imputation processing of multi-source time-series data after outlier removal: LSTM time-series interpolation is used to impute missing data caused by equipment maintenance or signal interruption, resulting in multi-source time-series data with imputed missing values. For data with missing durations not exceeding 30 minutes, the parameter values ​​for the missing period are predicted using an LSTM prediction model to ensure the integrity of the dataset. If the missing duration exceeds 30 minutes, manual verification and supplementation are used to avoid excessive time-series interpolation errors. Missing value imputation can employ simple imputation methods such as mean imputation, median imputation, and mode imputation; more complex interpolation methods such as linear interpolation and spline interpolation; and model-based methods such as K-nearest neighbor (KNN) imputation and regression imputation.

[0214] 3) Data normalization processing of the multi-source time series data after missing value imputation: The Min-Max normalization method is used to normalize the parameter data of all multi-source time series data after missing value imputation to the [0, 1] interval, eliminating the dimensional differences between different parameters (e.g., fluoride ion concentration is in mg / L, flow rate is in m³ / h), and avoiding the impact of dimensional differences on subsequent model training, resulting in normalized multi-source time series data. The normalization formula can be expressed as: Where x represents the original parameter value of the multi-source time series data after missing value completion processing. min x is the minimum value of this parameter. max This is the maximum value of the parameter;

[0215] 4) Perform multi-parameter time-series alignment processing on the normalized multi-source time-series data: Due to the differences in data acquisition frequencies among various detection devices, it is necessary to perform time-series alignment on the normalized multi-source time-series data to obtain the final standardized multi-source time-series data. For example, the acquisition timestamp of the influent flow rate can be used as a benchmark to calibrate and adjust the acquisition times of other parameters, ensuring that the data of each parameter matches one by one at the same time, thus forming a time-synchronized time-series dataset.

[0216] In this embodiment, by acquiring initial multi-source time-series data at the current moment, outlier removal, missing value completion, data normalization, and multi-parameter time-series alignment are performed on the initial multi-source time-series data to obtain standardized multi-source time-series data. This standardized multi-source time-series data is then used as input to the digital twin estimation model and the LSTM prediction model, significantly improving the quality and consistency of the input data. Because the models receive accurate, complete, and dimensionlessly consistent data, the digital twin estimation model can more accurately estimate the target crystallization kinetic sedimentation rate, and the LSTM prediction model can more accurately predict effluent indicators. Furthermore, based on the target crystallization kinetic sedimentation rate and predicted effluent indicators, the reinforcement learning decision model can calculate more reasonable and reliable target control parameters, thereby achieving precise control of the fluoride removal process in fluoride-containing wastewater crystallization. This data preprocessing mechanism ensures the effectiveness and stability of the system from the source, enabling the model-based control strategy to achieve its maximum efficiency.

[0217] Optionally, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: acquiring system feedback operation data of the defluoridation process of fluoride-containing wastewater crystallization within the next time window; and correcting the target control parameters based on the system feedback operation data and preset standard operation data.

[0218] For example, the system feedback operation data can be the actual operating data obtained by controlling the operation of the fluoride-containing wastewater crystallization and defluorination process in the next time window adjacent to the current time window based on the target control parameters, including but not limited to the effluent fluoride ion concentration, average crystal particle size, pH value of the water in the crystallization reaction unit, and stirring intensity.

[0219] Understandably, the system feedback data can accurately reflect the actual operating conditions and control effects of the process, and is an important basis for evaluating the effectiveness of the current control strategy.

[0220] For example, the preset standard operating data can be set as ideal operating conditions or a reasonable range of values ​​based on historical optimized operating data, process design parameters, or industry expert experience.

[0221] For example, in some embodiments, the system feedback operation data can be compared one by one with the preset standard operation data to calculate the corresponding deviation (such as percentage deviation), and the target control parameters can be further corrected in real time based on the deviation. For example, when the effluent fluoride concentration in the system feedback operation data reaches 8.4 mg / L, close to the upper limit of the discharge standard, the Ca / F molar ratio is adjusted from 0.8 to 0.84, a correction of 5%, so that the effluent fluoride concentration drops back to less than or equal to 8 mg / L; when the average crystal particle size decreases to 45 μm, the seed crystal dosage is adjusted from 80 mg / L to 88 mg / L, a correction of 10%, to promote normal crystal growth; in some embodiments, the system feedback operation data can be compared with the preset standard operation data one by one, and the corresponding difference (such as absolute difference or relative difference) can be calculated, and the target control parameters can be further adjusted in real time based on the difference; in some embodiments, statistical methods are used to compare the system feedback operation data with the preset standard operation data, and the corresponding root mean square deviation or standard deviation (measuring the overall deviation between the system feedback operation data and the preset standard operation data) can be calculated, and the target control parameters can be further adjusted in real time based on the root mean square deviation or standard deviation.

[0222] In one possible implementation, the parameter correction interval period can be set according to the degree of fluctuation in operating conditions. For example, the correction interval period can be set to 5 minutes to 10 minutes. For instance, when the operating conditions fluctuate greatly, the correction interval period is selected as 5 minutes, and when the operating conditions are relatively stable, the correction interval period is selected as 10 minutes.

[0223] In one possible implementation, when correcting the target control parameter, the magnitude of a single correction shall not exceed 10% of the previous control parameter, to avoid drastic fluctuations in system operating conditions caused by sudden parameter changes. For example, if the previous Ca / F molar ratio was 0.8, the maximum value range after this correction would be 0.88 and the minimum value would be 0.72.

[0224] This application embodiment introduces a real-time feedback correction mechanism. By acquiring system feedback operation data after the crystallization and defluorination process of fluoride-containing wastewater has run, and correcting the target control parameters based on the system feedback operation data and preset standard operation data, the system can dynamically adjust the target control parameters according to the system feedback operation data and preset standard operation data. This effectively copes with various uncertainties and disturbances in the actual crystallization and defluorination process, compensates for the limitations of model prediction, and ensures that the crystallization and defluorination process of fluoride-containing wastewater always operates in the optimal or near-optimal process state. This significantly improves the robustness, stability, and accuracy of control, and ultimately ensures the continuous compliance of effluent water quality and the optimization of operating efficiency.

[0225] It is understood that, through continuous iterative optimization of the LSTM prediction model and the reinforcement learning decision model, combined with real-time feedback correction of the target control parameters, the effluent fluoride concentration can be ensured to stably meet the discharge standards for fluoride-containing wastewater. This keeps the consumption of calcium salt reagents at a low level, while ensuring excellent crystal quality and stable system operation without significant fluctuations in operating conditions.

[0226] Optionally, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: identifying system operation faults based on a fault identification model and standardized multi-source time-series data; and triggering a preset emergency control plan and alarm prompts according to the fault type corresponding to the identified system operation fault in response to the identification of an operation fault.

[0227] For example, the fault identification model can be a fault identification and root cause localization model, which can be constructed using the random forest algorithm.

[0228] For example, fault types can be stored in a fault mode library, and fault types include, but are not limited to, deterioration of crystallization effect, abnormal loss of crystal seeds, reagent failure, and abnormal fluctuation of water pH.

[0229] In one possible implementation, based on the historical operation dataset of the intelligent control system for fluoride removal crystallization in fluoride-containing wastewater stored in the cloud, the system analyzes various common fault types that are prone to occur during actual operation. For each fault type, it extracts the corresponding characteristic parameters and their variation patterns. These characteristic parameters include, but are not limited to, system influent and effluent water quality parameters, crystal particle morphology and average crystal particle size distribution, real-time pH fluctuation data, calcium salt dosage, and crystallization reaction temperature. Each fault type is associated with its corresponding characteristic parameters and variation patterns and stored in a fault mode library.

[0230] For example, for each fault type, based on its corresponding characteristic parameters and variation patterns, an emergency control plan matching that fault type is pre-set, and the fault type, characteristic parameters and variation patterns, and the corresponding emergency control plan are associated and stored in a fault mode library. The emergency control plan corresponding to the fault type includes, but is not limited to, adjusting the dosage of reagents, controlling the amount of seed crystals added, and calibrating the pH value of the water.

[0231] For example, alarm notifications can take many forms, such as audible and visual alarms, pop-up notifications on the control interface, SMS notifications, emails, etc.

[0232] In this embodiment, one possible implementation is as follows: Standardized multi-source time-series data is input into the fault identification model for corresponding parameter feature analysis and matching to achieve real-time identification of the current system operation fault and accurately locate the root cause of the fault. After the system completes fault identification and root cause location, in response to the identification of the system operation fault, based on the fault mode library, according to the fault type corresponding to the operation fault, the preset emergency control plan corresponding to the fault type is determined, the corresponding preset emergency control plan is automatically triggered and executed, and the audible and visual alarm module is simultaneously activated to issue a clear alarm signal to remind the operator to intervene in time for verification and handling, so as to ensure that the fault is dealt with quickly.

[0233] This application embodiment introduces a fault identification and emergency response mechanism. Based on a fault identification model, it identifies system operation faults using standardized multi-source time-series data. When an operation fault is identified, it triggers a preset emergency control plan and alarm prompts according to the fault type. This further enhances the robustness and safety of the system, ensuring that even in the event of a sudden fault in a complex industrial environment, the system can respond quickly, effectively preventing the fault from escalating and ensuring the continuity, stability, and safe operation of the fluoride removal process in the crystallization process of fluoride-containing wastewater.

[0234] Optionally, the digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment further includes: storing all information such as system operating parameter data, fault identification results, implementation process of preset emergency control scheme, and control effect verification data in real time into a cloud server as sample data for fault identification model iteration, and periodically (preferably weekly) updating the fault mode library, simultaneously optimizing the emergency control scheme and random forest algorithm model parameters, continuously improving the system's anti-interference ability, identification accuracy, and emergency response efficiency for various faults, and further ensuring the stability and reliability of system operation.

[0235] In summary, the digital twin-driven intelligent control method and system for defluoridation of fluoride-containing wastewater through crystallization provided in this application has the following beneficial effects:

[0236] 1) Breaking through the traditional linear control mode, a system is constructed based on the precipitation rate of crystallization kinetics (optimal range). A progressive closed-loop control architecture with as its core; and a digital twin estimation model to accurately estimate the target crystallization kinetics precipitation rate (estimation error). The system outputs optimal control parameters based on a dual AI model consisting of an LSTM prediction model and a reinforcement learning decision model. After the control parameters are executed, a feedback correction mechanism is introduced. Combined with daily model self-learning, timed control parameter correction, and fault emergency parameter control, a closed-loop mechanism of "estimation-decision-control-correction-optimization" is formed, which effectively ensures the stable and efficient operation of the system and significantly improves the system control accuracy.

[0237] 2) Multiple vertically parallel microchannel reaction columns are used, and an intermittent, staggered operation mode is adopted to avoid the impact of concentrated operation load; the number and opening ratio of microchannel reaction columns can be dynamically adjusted according to the target crystallization kinetic precipitation rate range, and the effective reaction volume and supersaturation ratio can be adjusted without impact to achieve stable and precise control of crystallization rate.

[0238] 3) Construct a closed-loop seed crystal system of "separation-regeneration-reuse". Small crystal particles are separated, regenerated and reused through a hydrocyclone separator to stabilize the number of crystal nuclei, maintain seed crystal activity and improve overall crystal quality. At the same time, the amount of regenerated seed crystals added is linked and matched with the crystallization kinetics precipitation rate to precisely control the amount of seed crystals added and the opening of the addition valve, providing strong support for the stable and controllable crystallization kinetics precipitation rate.

[0239] 4) Build a dual AI collaborative architecture of "LSTM prediction model + reinforcement learning decision model" to provide algorithmic support for precise and intelligent control of crystallization kinetic precipitation rate; at the same time, build a fault identification model and fault mode library based on random forest algorithm to realize rapid identification of system faults, root cause location and emergency automatic parameter adjustment, ensure the stable and reliable operation of the progressive cyclic control system, and further improve the system's intelligence level and engineering industrial adaptability.

[0240] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the embodiments of the apparatus of the present invention, please refer to the embodiments of the method of the present invention.

[0241] Figure 4 This is a schematic diagram of the structure of a digital twin-driven intelligent control device for defluoridation of fluoride-containing wastewater crystallization, provided in an embodiment of this application. Figure 4 As shown, the intelligent control device 40 for defluorination of fluoride-containing wastewater crystallization includes an acquisition module 410, a first processing module 420, a second processing module 430, and a control module 440.

[0242] Among them, the acquisition module 410 is used to acquire standardized multi-source time-series data of the fluoride-containing wastewater crystallization and defluorination process running within the current time window;

[0243] The first processing module 420 is used to input standardized multi-source time series data into the digital twin estimation model and the LSTM prediction model respectively, to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model and the predicted effluent index output by the LSTM prediction model.

[0244] The second processing module 430 inputs the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model to calculate the control parameters and obtain the target control parameters output by the reinforcement learning decision model.

[0245] The control module 440 is used to control the crystallization and defluorination process of fluoride-containing wastewater in the next time window adjacent to the current time window according to the target adjustment parameters.

[0246] In one possible implementation, the digital twin estimation model includes an ion solubility product thermodynamic model, a crystallization kinetic model, and a time-series data correction model. The standardized multi-source time-series data includes the crystallization reaction temperature. The first processing module 420 is specifically used to: calculate the calcium fluoride solubility product corresponding to the crystallization reaction temperature based on the ion solubility product thermodynamic model; calculate the simulated crystallization kinetic precipitation rate based on the crystallization kinetic model, the standardized multi-source time-series data, and the calcium fluoride solubility product; and correct the simulated crystallization kinetic precipitation rate based on the time-series data correction model to obtain the target crystallization kinetic precipitation rate.

[0247] In one possible implementation, the standardized multi-source time-series data also includes influent parameters of fluoride-containing wastewater, calcium salt dosing parameters, crystallization reaction process parameters, and the current crystallization kinetic sedimentation rate. The first processing module 420 is further used to: calculate the current calcium ion concentration and current fluoride ion concentration in the crystallization reaction tank based on the crystallization kinetic model, according to the influent parameters of fluoride-containing wastewater, calcium salt dosing parameters, and the current crystallization kinetic sedimentation rate; calculate the supersaturation ratio based on the current calcium ion concentration, current fluoride ion concentration, and calcium fluoride solubility product; and calculate the simulated crystallization kinetic sedimentation rate based on the supersaturation ratio, crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetic model.

[0248] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a compensation module (not shown). This compensation module is used to: calculate the crystallization kinetic sedimentation rate deviation between the current crystallization kinetic sedimentation rate and the simulated crystallization kinetic sedimentation rate; use a PID control algorithm to calculate the compensation adjustment amount of the current growth rate compensation constant based on the crystallization kinetic sedimentation rate deviation; and accumulate the compensation adjustment amount to the current growth rate compensation constant to obtain the target growth rate compensation constant.

[0249] In one possible implementation, the first processing module 420 is further configured to: iteratively calculate the precipitation rate of the simulated crystallization kinetics based on the supersaturation ratio, crystallization reaction process parameters, and target growth rate compensation constant.

[0250] In one possible implementation, the second processing module 430 is specifically used to: input the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model, take maintaining the target crystallization kinetic sedimentation rate within the preset crystallization kinetic sedimentation rate range as the core control objective, and output the target control parameters in combination with the compliance constraints of the predicted effluent index.

[0251] In one possible implementation, the target control parameters include the Ca / F molar ratio, pH setpoint, seed dosage, stirring intensity, and the number of microchannel reaction columns opened. The control module 440 is also used to: determine the number of microchannel reaction columns opened and the spatial distribution pattern of the microchannel reaction columns based on the target crystallization kinetic precipitation rate, wherein the number of microchannel reaction columns opened is determined based on the total number of microchannel reaction columns, and the opened microchannel reaction columns are staggered at intervals; and determine the target opening degree of the seed addition control valve and the corresponding seed dosage based on the target crystallization kinetic precipitation rate, wherein the opening degree of the seed addition control valve is positively correlated with the seed dosage.

[0252] In one possible implementation, the LSTM prediction model is trained as follows: First, a first multi-source time-series training dataset is acquired, which is a preprocessed, standardized historical multi-source time-series dataset; the first multi-source time-series training dataset is input into the LSTM prediction model to be trained to predict water discharge indicators, resulting in predicted water discharge indicator values ​​output by the LSTM prediction model; the mean square error between the predicted water discharge indicator values ​​and the true values ​​of the water discharge indicators in the first multi-source time-series training dataset is used as the loss function to iteratively optimize the model parameters of the LSTM prediction model to be trained until the loss function converges, thus obtaining the LSTM prediction model.

[0253] In one possible implementation, the reinforcement learning decision model is trained as follows: acquiring second multi-source time-series training data; inputting the second multi-source time-series training data into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model to be trained; using the multi-weighted reward function as the optimization objective, iteratively optimizing the model parameters of the reinforcement learning decision model to be trained based on the control parameters until the control parameters output by the reinforcement learning decision model to be trained can stabilize the score of the multi-weighted reward function within a preset score range, thus obtaining the reinforcement learning decision model.

[0254] In one possible implementation, the multi-weighted reward function includes weights for effluent indicator compliance, calcium salt dosage optimization, crystal size, and system stability; the total score for these weights is 100 points.

[0255] In one possible implementation, the acquisition module 410 is specifically used to: acquire the initial multi-source time series data of the fluoride-containing wastewater crystallization defluorination process running within the current time window; preprocess the initial multi-source time series data to obtain standardized multi-source time series data, the preprocessing including outlier removal processing, missing value completion processing, data normalization processing and multi-parameter time series alignment processing.

[0256] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a correction module (not shown), which is used to: acquire system feedback operation data of the defluoridation process of fluoride-containing wastewater crystallization within the next time window; and correct the target control parameters based on the system feedback operation data and the preset standard operation data.

[0257] In one possible implementation, the intelligent control device for defluoridation of fluoride-containing wastewater crystallization also includes a fault identification module (not shown). This fault identification module is used to: identify system operation faults based on a fault identification model and standardized multi-source time-series data; and in response to the identification of an operation fault in the system, trigger a preset emergency control plan and alarm prompts according to the fault type corresponding to the operation fault.

[0258] The intelligent control device for defluoridation of fluoride-containing wastewater crystallization provided in this application embodiment can be used to execute the method steps of the above method embodiment. The specific implementation and technical effects are similar, and will not be repeated here.

[0259] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the electronic device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0260] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0261] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0262] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0263] The memory may include random access memory (RAM) and non-volatile memory (NVM), such as at least one disk storage device.

[0264] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0265] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0266] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0267] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0268] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0269] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0270] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0271] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0272] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0273] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0274] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A digital twin-driven intelligent control method for defluoridation of fluoride-containing wastewater through crystallization, characterized in that, include: Obtain standardized multi-source time-series data of the fluoride removal crystallization process in fluoride-containing wastewater within the current time window, wherein the standardized multi-source time-series data includes the crystallization reaction temperature; The standardized multi-source time-series data is input into a digital twin estimation model to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model. This includes: the digital twin estimation model comprising an ion solubility product thermodynamic model, a crystallization kinetic model, and a time-series data correction model; calculating the calcium fluoride solubility product corresponding to the crystallization reaction temperature based on the ion solubility product thermodynamic model; and calculating the simulated crystallization kinetic precipitation rate based on the crystallization kinetic model, using the standardized multi-source time-series data and the calcium fluoride solubility product. , This represents the precipitation rate of the simulated crystallization kinetics at the current time t. denoted by t, representing the current growth rate compensation constant at time t; g represents the growth stage; S(t) represents the supersaturation ratio at time t. This represents the total specific surface area of ​​the crystals at the current time t. , Indicates the standardized crystallization concentration; Indicates the effective volume of the crystallization reactor; This indicates the density of calcium fluoride; This represents the average crystal grain size; the growth rate compensation constant can be adjusted based on the compensation adjustment amount. , This represents the adjusted growth rate compensation constant corresponding to the current time t. This represents the compensation adjustment amount corresponding to the current time t. Represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. This represents the deviation between the current crystallization kinetic precipitation rate and the simulated crystallization kinetic precipitation rate at the current time t; the simulated crystallization kinetic precipitation rate is corrected based on the time-series data correction model to obtain the target crystallization kinetic precipitation rate; The standardized multi-source time series data is input into the LSTM prediction model to obtain the predicted water discharge index output by the LSTM prediction model. The target crystallization kinetic sedimentation rate and the predicted effluent index are input into the reinforcement learning decision model to calculate the control parameters, and the target control parameters output by the reinforcement learning decision model are obtained. Based on the target control parameters, control the fluoride removal process of fluoride-containing wastewater crystallization in the next time window adjacent to the current time window.

2. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to claim 1, characterized in that, The standardized multi-source time-series data also includes influent parameters for fluoride-containing wastewater, calcium salt dosage parameters, crystallization reaction process parameters, and the current crystallization kinetic precipitation rate. The calculation of the simulated crystallization kinetic precipitation rate based on the crystallization kinetic model, according to the standardized multi-source time-series data and the calcium fluoride solubility product, includes: Based on the crystallization kinetics model, the current calcium ion concentration and current fluoride ion concentration in the crystallization reaction tank are calculated according to the influent parameters of the fluoride-containing wastewater, the calcium salt reagent addition parameters, and the current crystallization kinetics precipitation rate. Calculate the supersaturation ratio based on the current calcium ion concentration, the current fluoride ion concentration, and the calcium fluoride solubility product. The simulated crystallization kinetics precipitation rate is calculated based on the supersaturation ratio, the crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetics model.

3. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to claim 2, characterized in that, Also includes: Calculate the crystallization kinetic precipitation rate deviation between the current crystallization kinetic precipitation rate and the simulated crystallization kinetic precipitation rate; Using a PID control algorithm, the compensation adjustment amount of the current growth rate compensation constant is calculated based on the precipitation rate deviation of the crystallization kinetics. The compensation adjustment amount is accumulated to the current growth rate compensation constant to obtain the target growth rate compensation constant.

4. The intelligent control method for crystallization-based fluoride removal from fluoride-containing wastewater according to claim 3, characterized in that, The step of calculating the simulated crystallization kinetic precipitation rate based on the supersaturation ratio, the crystallization reaction process parameters, and the current growth rate compensation constant in the crystallization kinetic model includes: The precipitation rate of the simulated crystallization kinetics is calculated iteratively based on the supersaturation ratio, the crystallization reaction process parameters, and the target growth rate compensation constant.

5. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to any one of claims 1 to 4, characterized in that, The step of inputting the target crystallization kinetic precipitation rate and the predicted effluent index into a reinforcement learning decision model to calculate the control parameters, and obtaining the target control parameters output by the reinforcement learning decision model, includes: The target crystallization kinetic sedimentation rate and the predicted effluent index are input into the reinforcement learning decision model. The core control objective is to maintain the target crystallization kinetic sedimentation rate within the preset crystallization kinetic sedimentation rate range. Combined with the compliance constraints of the predicted effluent index, the target control parameters are output.

6. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to claim 5, characterized in that, The target control parameters include The method further includes: molar ratio, pH setting, seed crystal dosage, stirring intensity, and the number of microchannel reaction columns opened; Based on the target crystallization kinetic precipitation rate, the number of microchannel reaction columns to be opened and the spatial distribution pattern of the microchannel reaction columns are determined. The number of microchannel reaction columns to be opened is determined based on the total number of microchannel reaction columns, and the opened microchannel reaction columns are staggered at intervals. Based on the target crystallization kinetic precipitation rate, the target opening degree of the seed addition control valve and the corresponding seed addition amount are determined, and the opening degree of the seed addition control valve is positively correlated with the seed addition amount.

7. The intelligent control method for crystallization-based fluoride removal from fluoride-containing wastewater according to any one of claims 1 to 4, characterized in that, The LSTM prediction model was trained in the following way: Obtain the first multi-source time series training data, which is a preprocessed historical standardized multi-source time series dataset. The first multi-source time series training data is input into the LSTM prediction model to be trained to predict the water discharge index, and the predicted value of the water discharge index output by the LSTM prediction model to be trained is obtained. Using the mean square error between the predicted values ​​of the effluent indicators and the true values ​​of the effluent indicators in the first multi-source time-series training data as the loss function, the model parameters of the LSTM prediction model to be trained are iteratively optimized until the loss function converges, thus obtaining the LSTM prediction model.

8. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to any one of claims 1 to 4, characterized in that, The reinforcement learning decision model is trained in the following way: Obtain the second multi-source time-series training data; The second multi-source time-series training data is input into the reinforcement learning decision model to be trained to calculate the control parameters, thereby obtaining the control parameters output by the reinforcement learning decision model to be trained. Using a multi-weighted reward function as the optimization objective, the model parameters of the reinforcement learning decision model to be trained are iteratively optimized according to the control parameters until the control parameters output by the reinforcement learning decision model to be trained can make the score of the multi-weighted reward function stabilize within a preset score range, thus obtaining the reinforcement learning decision model.

9. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to claim 8, characterized in that, The multi-weighted reward function includes weights for effluent quality compliance, calcium salt dosage optimization, crystal size, and system stability. The total score for the weighting of the effluent indicators, the weighting of the calcium salt dosage optimization, the weighting of the crystal size, and the weighting of the system stability is 100 points.

10. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to any one of claims 1 to 4, characterized in that, The acquisition of standardized multi-source time-series data of the fluoride removal crystallization process in fluoride-containing wastewater within the current time window includes: Acquire initial multi-source time-series data of the fluoride crystallization and defluorination process in fluoride-containing wastewater within the current time window; The initial multi-source time series data is preprocessed to obtain the standardized multi-source time series data. The preprocessing includes outlier removal, missing value completion, data normalization, and multi-parameter time series alignment.

11. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to any one of claims 1 to 4, characterized in that, Also includes: Obtain system feedback operation data for the crystallization and defluoridation process of fluoride-containing wastewater within the next time window; The target control parameters are adjusted based on the system feedback operation data and the preset standard operation data.

12. The intelligent control method for defluoridation of fluoride-containing wastewater by crystallization according to any one of claims 1 to 4, characterized in that, Also includes: Based on the fault identification model, system operation faults are identified according to the standardized multi-source time series data. In response to the detection of a system malfunction, a preset emergency control plan and alarm prompts are triggered according to the malfunction type.

13. A digital twin-driven intelligent control device for defluoridation of fluoride-containing wastewater crystallization, characterized in that, include: The acquisition module is used to acquire standardized multi-source time-series data of the crystallization and defluorination process of fluoride-containing wastewater within the current time window, wherein the standardized multi-source time-series data includes the crystallization reaction temperature; The first processing module is used for The standardized multi-source time-series data is input into a digital twin estimation model to obtain the target crystallization kinetic precipitation rate output by the digital twin estimation model. This includes: the digital twin estimation model comprising an ion solubility product thermodynamic model, a crystallization kinetic model, and a time-series data correction model; calculating the calcium fluoride solubility product corresponding to the crystallization reaction temperature based on the ion solubility product thermodynamic model; and calculating the simulated crystallization kinetic precipitation rate based on the crystallization kinetic model, using the standardized multi-source time-series data and the calcium fluoride solubility product. , This represents the precipitation rate of the simulated crystallization kinetics at the current time t. denoted by t, representing the current growth rate compensation constant at time t; g represents the growth stage; S(t) represents the supersaturation ratio at time t. This represents the total specific surface area of ​​the crystals at the current time t. , Indicates the standardized crystallization concentration; Indicates the effective volume of the crystallization reactor; This indicates the density of calcium fluoride; The growth rate compensation constant, which indicates the average crystal grain size, can be adjusted based on the compensation adjustment amount. , This represents the adjusted growth rate compensation constant corresponding to the current time t. This represents the compensation adjustment amount corresponding to the current time t. Represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. This represents the deviation between the current crystallization kinetic sedimentation rate and the simulated crystallization kinetic sedimentation rate at the current time t; the simulated crystallization kinetic sedimentation rate is corrected based on the time series data correction model to obtain the target crystallization kinetic sedimentation rate; the standardized multi-source time series data is input into the LSTM prediction model to obtain the predicted effluent index output by the LSTM prediction model; The second processing module is used to input the target crystallization kinetic sedimentation rate and the predicted effluent index into the reinforcement learning decision model to calculate the control parameters and obtain the target control parameters output by the reinforcement learning decision model. The control module is used to control the crystallization and defluorination process of fluoride-containing wastewater in the next time window adjacent to the current time window, according to the target adjustment parameters.

14. A digital twin-driven intelligent control system for defluoridation of fluoride-containing wastewater through crystallization, characterized in that, include: Intelligent control unit and seed regeneration and reuse unit; The intelligent control unit is used to execute the intelligent control method for defluoridation of fluoride-containing wastewater by crystallization as described in any one of claims 1 to 12; The seed regeneration and reuse unit is used to classify and separate the calcium fluoride crystal mixture generated by the above-mentioned intelligent control method for defluorination through crystallization of fluoride-containing wastewater, and to regenerate the seed crystals that do not meet the requirements for crystal reuse.

15. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the intelligent control method for crystallization and fluoride removal from fluoride-containing wastewater as described in any one of claims 1 to 12.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the intelligent control method for crystallization and defluorination of fluoride-containing wastewater as described in any one of claims 1 to 12.

17. A computer program product, characterized in that, include: A computer program, when executed by a processor, implements the intelligent control method for defluoridation of fluoride-containing wastewater by crystallization as described in any one of claims 1 to 12.

Images