Membrane pollution regulation and control method and system of water treatment system

Abstract

The invention belongs to the technical field of water treatment membranes, and discloses a membrane pollution regulation and control method and system for a water treatment system, and the method comprises the steps: constructing an LSTM pollution prediction model and a multi-scale membrane pollution model; acquiring working condition parameters of operation of the water treatment system in real time, and predicting membrane pollution parameters in a preset time period by using an LSTM pollution prediction model according to the working condition parameters; determining constraint conditions according to the multi-scale membrane pollution model, and constructing a grey wolf optimization algorithm based on the constraint conditions; and inputting the membrane pollution parameters into a grey wolf optimization algorithm to obtain membrane pollution regulation and control parameters, and determining a membrane pollution regulation and control strategy according to the membrane pollution regulation and control parameters. The method has the advantages that through the synergistic effect of the multi-scale membrane pollution model and the dynamically updated LSTM pollution prediction model, accurate prediction of the TMP is achieved, the prediction error is controllable, the membrane pollution regulation and control strategy can be determined 4-8 hours in advance, sufficient response time is provided for process adjustment, and rapid deterioration of the performance of a membrane system is effectively avoided.

Description

Technical Field

[0001] This invention discloses a method and system for controlling membrane fouling in a water treatment system, belonging to the field of water treatment membrane technology. Background Technology

[0002] Membrane separation technology has been widely used in water treatment due to its high efficiency in retaining pollutants. However, membrane fouling remains a key bottleneck restricting system performance. When pollutants such as organic matter, colloids, and microorganisms continuously deposit on the membrane surface to form a dense fouling layer, it will cause a cliff-like decline in membrane flux and a nonlinear increase in transmembrane pressure, directly leading to a surge in system energy consumption and shortened membrane module lifespan.

[0003] Current pollution control technologies in industrial applications face three core dilemmas:

[0004] First, traditional chemical cleaning strategies rely on fixed cycles or empirical thresholds, making it difficult to dynamically respond to changes in the fouling characteristics caused by fluctuations in influent water quality. They generally suffer from irreversible damage to membrane performance due to cleaning lag or waste of chemicals due to over-cleaning.

[0005] Secondly, although simple data-driven models (such as neural networks) can fit historical operating data, they lack in-depth analysis of physicochemical processes such as membrane fouling mass transfer mechanism and interfacial interaction. When water quality changes suddenly, the prediction accuracy drops sharply, causing the control strategy to fail.

[0006] Furthermore, existing control methods do not couple the multi-scale mechanisms of pollution formation—including the synergistic effects of nanoscale colloidal adsorption, micron-scale biofilm growth, and molecular-level concentration polarization—leading to a significant limitation on the universality of optimization strategies in complex water quality scenarios. Summary of the Invention

[0007] The purpose of this invention is to provide a method for controlling membrane fouling in a water treatment system to address the shortcomings of existing methods that rely on fixed cycles and are unable to dynamically respond to sudden changes in influent water quality. Furthermore, conventional data-driven neural network models suffer a sharp drop in prediction accuracy during water quality changes, making them unsuitable for controlling membrane fouling in complex water quality scenarios. To achieve the above objective, this invention proposes a method for controlling membrane fouling in a water treatment system, the specific solution of which is as follows:

[0008] First aspect:

[0009] A method for controlling membrane fouling in a water treatment system includes the following steps:

[0010] Step 1: Construct an LSTM fouling prediction model and a multi-scale membrane fouling model;

[0011] Step 2: Collect the operating parameters of the water treatment system in real time, and use the LSTM fouling prediction model to predict the membrane fouling parameters for a preset time period based on the operating parameters.

[0012] Step 3: Determine the constraints based on the multi-scale membrane fouling model, and construct the Grey Wolf optimization algorithm based on the constraints;

[0013] Step 4: Input the membrane fouling parameters into the Grey Wolf optimization algorithm to obtain membrane fouling control parameters, and determine the membrane fouling control strategy based on the membrane fouling control parameters.

[0014] Preferably, step 4 is followed by:

[0015] Retrieve the membrane fouling prediction values ​​at a preset time point predicted by the LSTM fouling prediction model, and collect the actual membrane fouling values ​​at that preset time point;

[0016] Based on the relative deviation between the actual and predicted membrane fouling values, the weights of the LSTM fouling prediction model and the mutation probability of the Grey Wolf optimization algorithm are adjusted.

[0017] Preferably, a multi-scale membrane fouling model is constructed, including:

[0018] Step 1.1: Construct a membrane adsorption model of the microlayer based on the extended DLVO theory;

[0019] Step 1.2: Based on the Blatt model and the pore network model, construct a membrane pore fouling evolution model for the mesoscopic layer based on the membrane adsorption model;

[0020] Step 1.3: Construct a transmembrane pressure difference model for the macroscopic layer and couple it with the membrane pore fouling evolution model to form a multi-scale membrane fouling model.

[0021] Preferably, an LSTM pollution prediction model is constructed, including:

[0022] Obtain historical operating parameters of the water treatment system;

[0023] The water quality parameters and operating parameters from the historical operating conditions parameters are used as inputs;

[0024] Using the membrane state parameters from the historical operating conditions as labels, an LSTM pollution prediction model is trained.

[0025] Preferably, after step 2 and before step 3, the method further includes:

[0026] Determine whether the membrane fouling parameters have reached a preset threshold;

[0027] If not, the membrane fouling control strategy shall be determined based on the predetermined basic control parameters;

[0028] If so, the membrane fouling parameters are input into the Gray Wolf optimization algorithm to obtain membrane fouling control parameters, and the membrane fouling control strategy is determined based on the membrane fouling control parameters.

[0029] Preferably, step 3 includes the following steps:

[0030] The membrane fouling level is determined based on the membrane fouling parameters that reach the preset threshold.

[0031] Based on the membrane fouling level, a corresponding membrane fouling adjustment framework is determined. Accordingly, the Grey Wolf optimization algorithm generates membrane fouling control parameters based on the membrane fouling parameters and the membrane fouling adjustment framework.

[0032] Preferably, the membrane fouling parameters include: the transmembrane pressure difference growth rate and the membrane fouling layer growth rate.

[0033] Preferably, step 3 includes:

[0034] The membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold are determined based on the multi-scale membrane fouling mechanism model.

[0035] With the membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold as constraints, and with the optimization objectives of flux retention rate not being less than a first preset value, transmembrane pressure difference growth rate not being greater than a second preset value, and cleaning energy consumption not being greater than a third preset value, a gray wolf optimization algorithm is constructed.

[0036] Preferably, the gray wolf optimization algorithm is a gray wolf optimization algorithm that incorporates an adaptive mutation mechanism.

[0037] The second aspect:

[0038] A membrane fouling dynamic control system for a water treatment system, comprising:

[0039] The online monitoring module is used to collect the operating parameters of the water treatment system, including water quality parameters, operating parameters, and membrane state parameters.

[0040] The strategy generation unit is connected to the online monitoring module and has a built-in algorithm program for the membrane fouling control method described in the first aspect, which is used to generate a membrane fouling control strategy based on the received operating parameters.

[0041] An execution control module, connected to the strategy generation unit, is used to automatically adjust the operating parameters of the membrane system according to the received membrane fouling control strategy.

[0042] Beneficial effects: Through the synergistic effect of the multi-scale membrane fouling model and the dynamically updated LSTM fouling prediction model, the transmembrane pressure difference can be accurately predicted with the prediction error controlled within ±8%. Furthermore, the membrane fouling control strategy can be determined 4-8 hours in advance, providing sufficient response time for process adjustments and effectively avoiding the rapid deterioration of membrane system performance.

[0043] This invention optimizes the control parameters of the membrane through a control strategy, which can maintain the membrane flux rate above the preset value for a long period of time, extend the chemical cleaning cycle, significantly reduce operating energy consumption and maintenance costs, and extend the service life of the membrane module.

[0044] This invention employs long-term online data calibration and updates the model based on the deviation between predicted and actual values. This enables the model to adapt to different water quality conditions and maintain a stable control effect even under conditions of large fluctuations in influent water quality and complex operating conditions, ensuring long-term reliable operation of the system. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0047] Example 1:

[0048] To address the common problem of insufficient precision in membrane fouling control strategies in water treatment systems, this embodiment constructs a method for regulating membrane fouling in water treatment systems, such as... Figure 1 As shown, the method includes the following steps:

[0049] Step 1: Construct an LSTM fouling prediction model and a multi-scale membrane fouling model;

[0050] Among them, the construction of a multi-scale membrane fouling mechanism model includes:

[0051] Step 1.1: Construct a membrane adsorption model of the microlayer based on the extended DLVO theory;

[0052] Specifically, at the microscale, the extended DLVO theory is used to quantify the interaction energy between pollutants and the membrane surface. This model introduces an ionic strength correction term to specifically characterize the enhancing effect of calcium ions on adsorption energy. The total adsorption free energy of this model consists of the following components:

[0053] (1)

[0054] In the formula:

[0055] Represents the total interface free energy;

[0056] This represents the van der Waals interaction energy, characterizing the short-range intermolecular attraction;

[0057] It represents the electrostatic interaction energy, reflecting the electrostatic repulsion between the charged film surface and the charged pollutants;

[0058] It represents the steric hindrance energy, reflecting the conformational entropy change when polymeric pollutants are adsorbed on the membrane surface;

[0059] It should be noted that: (1) In the formula The quantitative calculation formula is shown in Equation 1.1 below:

[0060] (1.1)

[0061] In equation (1.1):

[0062] This represents the Hamaker constant (for the polyamide film-humic acid system). Membrane-colloid system );

[0063] These represent the equivalent radius of membrane surface roughness (typical value 10-50nm) and the particle size of pollutants (e.g., colloids 50-500nm, viruses 20-100nm), respectively.

[0064] This indicates the membrane-pollutant spacing (critical adsorption distance is taken as 1-3 nm).

[0065] (1) In the formula, The quantitative calculation formula is shown in equation (1.2):

[0066] (1.2)

[0067] In equation (1.2):

[0068] , representing the vacuum permittivity;

[0069] This represents the relative permittivity of water at 25°C.

[0070] This represents the reciprocal of the Debye length, where Ionic strength, in mol / L. Characterize the surface potential of the membrane and the pollutants.

[0071] (1) In the formula, The quantitative calculation formula is shown in equation (1.3):

[0072] (1.3)

[0073] In equation (1.3):

[0074] Represents Avogadro's constant;

[0075] This indicates the volume fraction of the solvent in the polymer;

[0076] This indicates the volume fraction of the polymer in the adsorption layer;

[0077] This indicates absolute temperature, at room temperature (25℃). Take 298K;

[0078] This represents the Boltzmann constant, approximately 1.38 × 10⁻⁶. -23 J / K;

[0079] The characteristic energy representing the thermal motion of molecules;

[0080] Indicates molecular weight.

[0081] Step 1.2: Based on the Blatt model and the pore network model, and based on the membrane adsorption model, construct a membrane pore fouling evolution model for the mesoscopic layer. The membrane pore fouling evolution model includes a dynamic evolution model of membrane porosity and a membrane fouling layer thickness evolution model.

[0082] Specifically, at the mesoscale, based on the deposition coefficient output by the membrane adsorption model of the microlayer, and combined with the Blatt concentration polarization model and the pore network model, the membrane porosity is constructed. The dynamic evolution model has the following evolution formula as shown in equation (21):

[0083] (twenty one)

[0084] (21) Where:

[0085] This indicates the initial porosity of the cleaning membrane surface;

[0086] This represents the deposition coefficient, which is related to the type of contaminant and is also calculated from the interfacial energy by the microlayer DLVO model. Quantitative determination: when , Take 0.02-0.05; when , Take a value of 0.05-0.1;

[0087] Indicates membrane flux;

[0088] This indicates the concentration of contaminants in the main solution (mg / L).

[0089] The formula for the evolution model of membrane fouling layer thickness is shown in equation (22) below:

[0090] (twenty two)

[0091] In the formula:

[0092] Indicates the initial contamination layer thickness;

[0093] This indicates the density of pollutant accumulation.

[0094] Step 1.3: Construct a transmembrane pressure difference model for the macroscopic layer and couple it with the membrane pore fouling evolution model to construct a multi-scale membrane fouling model.

[0095] Specifically, at the macroscopic scale, a transmembrane pressure difference model is constructed, which is coupled with a membrane pore fouling evolution model. Specifically, the expression for the transmembrane pressure difference model is shown in equation (3):

[0096] (3)

[0097] (3) Where:

[0098] This represents the initial transmembrane pressure difference (kPa).

[0099] This represents the viscosity of the fluid (Pa·s; the viscosity of water at 25°C is 0.001 Pa·s).

[0100] Indicates the intrinsic resistance of the membrane;

[0101] Indicates membrane fouling layer resistance ( ).

[0102] in The calculation formula is shown in Equation (3.1) below. This formula couples the membrane porosity dynamic evolution model and the membrane fouling layer thickness evolution model of the mesoscopic layer membrane pore fouling evolution model, as detailed below:

[0103] (3.1)

[0104] In the formula:

[0105] This indicates the average particle size of the pollutant (μm).

[0106] The multi-scale membrane fouling model constructed in this invention achieves cross-scale coupling from micro to macro: First, the microscopic molecular interaction energy is converted into a mesoscopic deposition coefficient through a membrane adsorption model; then, based on this deposition coefficient, a membrane pore fouling evolution model quantitatively describes the process of porosity decrease and fouling layer thickening caused by pollutant deposition; finally, the transmembrane pressure difference model accurately calculates the system resistance change caused by membrane fouling and the resulting macroscopic transmembrane pressure difference anomaly by coupling porosity and fouling layer thickness parameters.

[0107] This model comprehensively characterizes the transmission chain from microscopic adsorption energy to mesoscopic deposition behavior and then to macroscopic operating parameters. Specifically, at the microscopic level, it quantifies the interaction energy between pollutants and the membrane surface using DLVO theory; at the mesoscopic level, it dynamically calculates porosity based on the deposition coefficient. and the thickness of the contamination layer At the macro level, through pollution resistance With transmembrane pressure difference The quantitative relationship is used to map the pollution state to operating parameters.

[0108] Preferably, an LSTM pollution prediction model is constructed, including:

[0109] Obtain historical operating parameters of the water treatment system;

[0110] The water quality parameters and operating parameters from the historical operating conditions parameters are used as inputs;

[0111] An LSTM pollution prediction model was trained using membrane state parameters from historical operating conditions as labels.

[0112] Specifically, in this embodiment, constructing the LSTM pollution prediction model includes continuously collecting historical operating parameters of the water treatment system for 6 months, and randomly dividing the data into a training set and a validation set at a ratio of 7:3. The historical operating parameters include water quality parameters, operating parameters, and membrane state parameters. The LSTM pollution prediction model is trained using water quality parameters and operating parameters as inputs and membrane state parameters as labels.

[0113] Specifically, the network architecture is constructed before training the model. This includes building an LSTM neural network with an input layer, hidden layers, and an output layer. The input layer has 12 neurons that receive time-series data with 12-dimensional parameters. The first LSTM unit in the hidden layer has 32 neurons, and the second LSTM unit has 16 neurons. The output layer has 2 neurons (linearly activated) that output membrane pollution parameters for a future preset time period.

[0114] In this embodiment, the Adam optimizer is used (initial learning rate 0.001, loss function MSE≤0.001, number of iterations 200 rounds). After inputting the operating parameters, the model outputs the predicted value (membrane fouling parameters for a future preset period).

[0115] Step 2: Collect the operating parameters of the water treatment system in real time, and use the LSTM fouling prediction model to predict the membrane fouling parameters for a preset time period based on the operating parameters.

[0116] Specifically, after training and obtaining the LSTM pollution prediction model, operating parameters are input into the input layer, specifically the water quality parameters, operating parameters, and membrane state parameters of the membrane system in the preceding 12 hours. In this embodiment, the water quality parameters, operating parameters, and membrane state parameters are as follows:

[0117] Water quality parameters include: influent COD, turbidity, conductivity, microbial concentration, zeta potential, Concentration, pH value;

[0118] Operating parameters include: operating pressure and crossflow velocity;

[0119] Membrane state parameters include: transmembrane pressure difference (TMP), membrane flux, and temperature;

[0120] Align the above 12-dimensional parameters according to the time series to form the input matrix X∈R 12×12 (12 parameters × 12 hours).

[0121] The input matrix, composed of the current 12-hour operating parameters obtained above, is input into the LSTM fouling prediction model, and the model outputs the predicted membrane fouling parameters for the future preset period.

[0122] Furthermore, the membrane fouling parameters include: the transmembrane pressure difference growth rate and the membrane fouling layer growth rate.

[0123] Specifically, in this embodiment, the future preset time period is the next 4-8 hours, and the membrane fouling parameters are the transmembrane pressure difference growth rate (TMP growth rate) and the membrane fouling layer growth rate (fouling layer thickness growth rate).

[0124] To further improve the model's operating efficiency, this embodiment of the invention also includes the following steps after step 2 and before step 3:

[0125] Determine whether the membrane fouling parameters have reached a preset threshold;

[0126] If not, the membrane fouling control strategy shall be determined based on the predetermined basic control parameters;

[0127] Specifically, when the membrane fouling parameter (TMP growth rate) predicted by the LSTM fouling prediction model for the next 4-8 hours is not higher than a preset threshold, specifically 0.03 kPa / h in this embodiment, it is determined to be low-risk fouling, and the predetermined basic control parameters can be directly output to determine the membrane fouling control strategy.

[0128] If so, then step 3 is used to determine the membrane fouling control strategy.

[0129] Specifically, when the membrane fouling parameter (TMP growth rate) predicted by the LSTM fouling prediction model for the next 4-8 hours is higher than the preset threshold (i.e., TMP growth rate > 0.03 kPa / h), it is determined to be high-risk fouling, and step 3 is continued.

[0130] To further improve efficiency, the process after identifying high-risk pollution and before proceeding to step 3 includes:

[0131] The membrane fouling level is determined based on the membrane fouling parameters that reach the preset threshold.

[0132] Specifically, in this embodiment, when 0.03 kPa / h < TMP growth rate < 0.05 kPa / h, the membrane fouling level is determined to be the first high-risk fouling; when the TMP growth rate is ≥ 0.05 kPa / h, the membrane fouling level is determined to be the second high-risk fouling.

[0133] Based on the membrane fouling level, a corresponding membrane fouling adjustment framework is determined. Accordingly, the Grey Wolf optimization algorithm generates membrane fouling control parameters based on the membrane fouling parameters and the membrane fouling adjustment framework.

[0134] Specifically, when a membrane is identified as having the highest risk of fouling, the corresponding first-level membrane fouling framework is defined as follows: operating pressure reduced by 0.1 MPa, cross-flow velocity increased by 0.2-0.3 m / s, citric acid added at 20-50 mg / L, combined with pulse cleaning (30 s / cycle, 2 h interval). When a membrane is identified as having the second highest risk of fouling, the corresponding second-level membrane fouling framework is defined as follows: operating pressure reduced by 0.2 MPa, cross-flow velocity increased by 0.3 m / s, sodium hypochlorite added at 100-300 mg / L, and shutdown followed by 30 min of circulating cleaning. Within this framework, the Grey Wolf optimization algorithm generates precise membrane fouling control parameters based on the specific membrane fouling parameters.

[0135] Step 3: Determine the constraints based on the multi-scale membrane fouling model, and construct the Grey Wolf optimization algorithm based on the constraints;

[0136] Specifically, based on the constraints determined by the multi-scale membrane fouling model, the Grey Wolf Optimization Algorithm (GWO algorithm) is constructed. The optimization objective of the Grey Wolf Optimization Algorithm is shown in the following equation (4):

[0137] (4)

[0138] (4) Where:

[0139] Indicates membrane flux retention rate;

[0140] Indicates the TMP growth rate;

[0141] This indicates the energy consumption for cleaning.

[0142] Among them, the constraints of the optimization objective .

[0143] Furthermore, the weighting coefficients reflect the priorities of flux stability (40%), pollution control (40%), and energy consumption control (20%). Further, step 3 includes:

[0144] The membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold are determined based on the multi-scale membrane fouling mechanism model.

[0145] With the membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold as constraints, and with the optimization objectives of flux retention rate not being less than a first preset value, transmembrane pressure difference growth rate not being greater than a second preset value, and cleaning energy consumption not being greater than a third preset value, a gray wolf optimization algorithm is constructed.

[0146] Step 4: Input the membrane fouling parameters into the Grey Wolf optimization algorithm to obtain membrane fouling control parameters, and determine the membrane fouling control strategy based on the membrane fouling control parameters.

[0147] Specifically, membrane fouling parameters are input into the Grey Wolf optimization algorithm, which generates candidate solutions. Within the corresponding membrane fouling adjustment framework, the Grey Wolf algorithm randomly generates multiple sets of control parameters: operating pressure p, crossflow velocity v, cleaning agent concentration c, and flushing time t. The multi-scale membrane fouling model rapidly verifies each set of parameters based on physical laws. Specifically, control parameters that would lead to irreversible adsorption of membrane pollutants are eliminated if the adsorption energy barrier is <5KT. Also, control parameters that would cause severe membrane pore blockage are eliminated if the porosity is <0.5 within 12 hours. This avoids wasting computational resources on physically infeasible solutions. The objective function corresponding to the control parameters selected by the multi-scale membrane fouling model is calculated by the Grey Wolf algorithm. Select The minimum corresponding control parameter is determined, and then a specific control scheme is determined based on this control parameter.

[0148] It is important to note that the control parameters and their physical boundaries are defined as follows: operating pressure p, which affects membrane flux and fouling layer compaction; crossflow velocity v, which controls boundary layer thickness and shear force; chemical cleaning agent concentration c, which balances cleaning effect and cost; and rinsing time t, which determines the intensity of physical cleaning.

[0149] Furthermore, in order to improve the global search capability and convergence efficiency of the gray wolf optimization algorithm, the gray wolf optimization algorithm is a gray wolf optimization algorithm that introduces an adaptive mutation mechanism.

[0150] Specifically, population diversity indicators are introduced. (Defined as the ratio of the mean Euclidean distance between the current population and the initial population), based on Value adaptively and dynamically adjusts mutation probability :

[0151] (4.1)

[0152] In the formula:

[0153] Indicates the current iteration number;

[0154] This indicates the maximum number of iterations.

[0155] The adaptive mutation mechanism enhances global exploration, avoids getting trapped in local optima, focuses on local development, improves convergence accuracy, and achieves adaptive adjustment of strong mutation in the early stage and convergence in the later stage.

[0156] Furthermore, to improve the long-term prediction accuracy of the model, the prediction model is further calibrated over a long period. Step 4 includes the following:

[0157] Retrieve the membrane fouling prediction values ​​at a preset time point predicted by the LSTM fouling prediction model, and collect the actual membrane fouling values ​​at that preset time point;

[0158] Based on the relative deviation between the actual and predicted membrane fouling values, the weights of the LSTM fouling prediction model and the mutation probability of the Grey Wolf optimization algorithm are adjusted.

[0159] Specifically, the system retrieves predicted data by calling the membrane fouling prediction values ​​at preset time points (e.g., several hours later) output by the LSTM fouling prediction model, such as the growth trend of transmembrane pressure difference (TMP) or membrane flux decay rate. Actual data is collected in real time through the online monitoring module, acquiring actual membrane fouling values ​​at the same time point (e.g., TMP measured by differential pressure sensors, flux measured by mass flow meters).

[0160] Relative deviation calculation compares predicted and actual values ​​to determine the relative deviation (e.g., percentage of absolute error or root mean square error), evaluating the model's prediction accuracy. If the deviation exceeds a set threshold (e.g., 10%), the backpropagation (BP) algorithm is used to dynamically update the weight parameters of the LSTM network, prioritizing optimization of time-series features with significant deviations (e.g., sudden increases in turbidity or pressure fluctuations). An online learning mechanism incorporates the latest monitoring data into the training set, enhancing the model's adaptability to changes in operating conditions.

[0161] Furthermore, the mutation probability of the Grey Wolf optimization algorithm is adjusted according to the magnitude of the deviation. When the deviation is high, the mutation probability is increased to expand the parameter search range; when the deviation is low, the mutation probability is decreased to improve the convergence speed. Specifically, for every 5% increase in deviation, the mutation probability increases by 0.05, making it more closely reflect the actual evolution of membrane fouling.

[0162] The above adjustment process is executed automatically in a periodic manner (such as 30 minutes in this embodiment) or by event triggering (such as deviation exceeding the limit), forming a closed loop of "prediction-measurement-calibration". The corrected model is re-introduced for prediction, and the deviation change in the next period is recorded until the prediction accuracy stabilizes within the allowable range.

[0163] Example 2:

[0164] A membrane fouling dynamic control system for a water treatment system, comprising:

[0165] The online monitoring module is used to collect the operating parameters of the water treatment system, including water quality parameters, operating parameters, and membrane state parameters.

[0166] Specifically, the online monitoring devices for collecting water quality parameters include: an online COD analyzer for collecting influent COD, a laser scattering turbidimeter for collecting turbidity, a conductivity meter for collecting conductivity, an ATP detector for collecting microbial concentration, a Zeta potential analyzer for collecting zeta potential, and other devices for collecting... An ion chromatograph for high concentration and a pH meter for collecting pH values;

[0167] Online monitoring devices for acquiring operating parameters include: pressure transmitters for acquiring operating pressure and electromagnetic flowmeters for acquiring cross-flow velocity.

[0168] Online monitoring devices for acquiring membrane state parameters include: a differential pressure sensor for acquiring transmembrane pressure difference (TMP), a mass flow meter for acquiring membrane flux, and a platinum resistance thermometer for acquiring temperature.

[0169] The strategy generation unit is connected to the online monitoring module and has a built-in algorithm program for the membrane fouling control method described in Embodiment 1, which is used to generate a membrane fouling control strategy based on the received operating parameters.

[0170] Specifically, the strategy generation unit integrates an algorithm program based on the membrane fouling control method of Embodiment 1. Specifically, the algorithm program adopts the membrane fouling control method of the MCM-GWO-LSTM framework, and the MCM-GWO-LSTM framework executes the membrane fouling control method to generate a membrane fouling control strategy.

[0171] An execution control module, connected to the strategy generation unit, is used to automatically adjust the operating parameters of the membrane system according to the received membrane fouling control strategy.

[0172] Specifically, pressure, flow rate, cleaning frequency, and chemical dosage are dynamically adjusted through actuators such as frequency converters, electric valves, and chemical dosing devices.

[0173] In a further embodiment, the execution control module is also connected to the online monitoring module to provide real-time feedback on the execution effect to the online monitoring module, forming a closed loop of "monitoring-decision-execution-feedback" to ensure the accuracy of regulation.

[0174] In addition, in practical applications, an anomaly handling module is set up, which triggers an emergency control mode (such as enhanced backwashing or shutdown protection) when the monitored parameters exceed the threshold or the algorithm predicts an increased risk of pollution.

[0175] The following two examples of practical applications of this invention illustrate its effects in detail:

[0176] Example 3: Intelligent Control of RO Membrane Fouling in Industrial Wastewater Treatment. In an industrial wastewater treatment scenario, for influent conditions with COD concentrations of 80-120 mg / L and oil pollutant concentrations of 5-10 mg / L, the system operates at an operating pressure of 1.0 MPa and a crossflow velocity of 1.2 m / s. This invention achieves optimized control through the following technical solutions: Based on DLVO theory and combined with the oil droplet Zeta potential correction value, the interaction energy between the oil droplet and the membrane interface is accurately calculated; the Blatt model is used to dynamically simulate the evolution of the fouling layer porosity. The parameters of the Gray Wolf optimization algorithm are set (population size 40, iteration count 80), and mutation probability is implemented. The dynamic adjustment strategy (0.25→0.35→0.2) is employed. The LSTM model takes 12 hours of historical data (COD, oil content, conductivity) as input and outputs a 4-hour transmembrane pressure difference (TMP) prediction. When the LSTM predicts a TMP growth rate of 0.04 kPa / h (moderate pollution), the system automatically optimizes its operating parameters.

[0177] The crossflow velocity was increased to 1.5 m / s; 50 mg / L citric acid cleaning agent was added. After implementation, the flux decay rate decreased by 28%, the cleaning cycle was extended from 72 hours to 92 hours, and system energy consumption decreased by 20%.

[0178] Example 4: UF membrane fouling control in municipal wastewater. In the ultrafiltration system of a wastewater treatment plant, the focus is on the secondary biological effluent (turbidity 5-10 NTU, microbial concentration...). Under the operating conditions, the system achieves the following optimizations: An extended Sourirajan model is used to simulate biofilm growth dynamics; ATP concentration is introduced as a quantitative indicator of microbial activity. The backwashing intensity is optimized from... Upgraded to The sodium hypochlorite dosage was increased from 2 mg / L to 3.2 mg / L. The TMP rise rate was significantly reduced from 0.2 kPa / d to 0.11 kPa / d; membrane lifespan was extended from 12 months to 18 months; and chemical reagent consumption was reduced by 25%.

[0179] This invention constructs a self-learning intelligent control system for membrane fouling through the synergistic optimization of mechanistic and data models. Its technical advantages are specifically reflected in the following aspects:

[0180] By combining a multi-scale membrane fouling model with a dynamically updated LSTM fouling prediction model, accurate prediction of transmembrane pressure difference (TMP) can be achieved with a prediction error controlled within ±8%. Furthermore, membrane fouling control strategies can be determined 4-8 hours in advance, providing sufficient response time for process adjustments and effectively preventing a sharp deterioration in membrane system performance.

[0181] This invention optimizes the control parameters of the membrane through a control strategy, which can maintain the membrane flux rate above the preset value for a long period of time, extend the chemical cleaning cycle, significantly reduce operating energy consumption and maintenance costs, and extend the service life of the membrane module.

[0182] This invention employs long-term online data calibration and updates the model based on the deviation between predicted and actual values. This enables the model to adapt to different water quality conditions and maintain a stable control effect even under conditions of large fluctuations in influent water quality and complex operating conditions, ensuring long-term reliable operation of the system.

[0183] The above description is merely a few embodiments of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any modifications or alterations made by those skilled in the art without departing from the scope of the technical solution of the present invention using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for controlling membrane fouling in a water treatment system, characterized in that, Includes the following steps: Step 1: Construct an LSTM fouling prediction model and a multi-scale membrane fouling model; Step 2: Collect the operating parameters of the water treatment system in real time, and use the LSTM fouling prediction model to predict the membrane fouling parameters for a preset time period based on the operating parameters. Step 3: Determine the constraints based on the multi-scale membrane fouling model, and construct the Grey Wolf optimization algorithm based on the constraints; Step 4: Input the membrane fouling parameters into the Grey Wolf optimization algorithm to obtain membrane fouling control parameters, and determine the membrane fouling control strategy based on the membrane fouling control parameters.

2. The membrane fouling control method according to claim 1, characterized in that, Step 4 is followed by: Retrieve the membrane fouling prediction values ​​at a preset time point predicted by the LSTM fouling prediction model, and collect the actual membrane fouling values ​​at that preset time point; Based on the relative deviation between the actual and predicted membrane fouling values, the weights of the LSTM fouling prediction model and the mutation probability of the Grey Wolf optimization algorithm are adjusted.

3. The membrane fouling control method according to claim 1, characterized in that, Constructing a multi-scale membrane fouling model, including: Step 1.1: Construct a membrane adsorption model of the microlayer based on the extended DLVO theory; Step 1.2: Based on the Blatt model and the pore network model, construct a membrane pore fouling evolution model for the mesoscopic layer based on the membrane adsorption model; Step 1.3: Construct a transmembrane pressure difference model for the macroscopic layer and couple it with the membrane pore fouling evolution model to form a multi-scale membrane fouling model.

4. The membrane fouling control method according to claim 1, characterized in that, Constructing an LSTM pollution prediction model includes: Obtain historical operating parameters of the water treatment system; The water quality parameters and operating parameters from the historical operating conditions parameters are used as inputs; Using the membrane state parameters from the historical operating conditions as labels, an LSTM pollution prediction model is trained.

5. The membrane fouling control method according to claim 1, characterized in that, After step 2 and before step 3, the following are also included: Determine whether the membrane fouling parameters have reached a preset threshold; If not, the membrane fouling control strategy shall be determined based on the predetermined basic control parameters; If so, the membrane fouling parameters are input into the Gray Wolf optimization algorithm to obtain membrane fouling control parameters, and the membrane fouling control strategy is determined based on the membrane fouling control parameters.

6. The membrane fouling control method according to claim 5, characterized in that, Before step 3, the following is included: The membrane fouling level is determined based on the membrane fouling parameters that reach the preset threshold. Based on the membrane fouling level, a corresponding membrane fouling adjustment framework is determined. Accordingly, the Grey Wolf optimization algorithm generates membrane fouling control parameters based on the membrane fouling parameters and the membrane fouling adjustment framework.

7. The membrane fouling control method according to claim 1, characterized in that, The membrane fouling parameters include: the transmembrane pressure difference growth rate and the membrane fouling layer growth rate.

8. The membrane fouling control method according to claim 1, characterized in that, Step 3 includes: The membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold are determined based on the multi-scale membrane fouling mechanism model. With the membrane adsorption energy barrier threshold and the membrane fouling layer porosity threshold as constraints, and with the optimization objectives of flux retention rate not being less than a first preset value, transmembrane pressure difference growth rate not being greater than a second preset value, and cleaning energy consumption not being greater than a third preset value, a gray wolf optimization algorithm is constructed.

9. The membrane fouling control method according to claim 8, characterized in that, The gray wolf optimization algorithm is a gray wolf optimization algorithm that introduces an adaptive mutation mechanism.

10. A membrane fouling dynamic control system for a water treatment system, characterized in that, include: The online monitoring module is used to collect the operating parameters of the water treatment system, including water quality parameters, operating parameters, and membrane state parameters. The strategy generation unit is connected to the online monitoring module and has a built-in algorithm program for the membrane fouling control method according to any one of claims 1-9, which is used to generate a membrane fouling control strategy based on the received operating parameters. An execution control module, connected to the strategy generation unit, is used to automatically adjust the operating parameters of the membrane system according to the received membrane fouling control strategy.

Images