Membrane pollution state determination method based on multi-dimensional operation parameter association

Abstract

The invention relates to the technical field of membrane filtration, in particular to a membrane pollution state determination method based on multi-dimensional operation parameter association. The membrane pollution state determination method based on multi-dimensional operation parameter association comprises the following steps: acquiring multi-dimensional input parameters of an ultrafiltration membrane system in a current operation state; and preprocessing the obtained multi-dimensional input parameters to obtain a multi-dimensional parameter data set. And inputting the multi-dimensional parameter data set into a pre-established optimal analysis model to obtain a current membrane pollution index PI. Comparing the current membrane pollution index PI with each preset pollution grade range, quantitatively judging to obtain the pollution state grade of the ultrafiltration membrane, and determining treatment measures according to the pollution state grade. The method has the technical effects and advantages that the problem of single historical working condition data is solved, the data preprocessing process is perfected, the model design is reasonable, the adaptability is high, the operation is simple and convenient, the practicability is high, and the service life of the ultrafiltration membrane is prolonged.

Description

Technical Field

[0001] This invention relates to the field of membrane filtration technology, and more specifically to a method for determining membrane fouling status based on multidimensional operating parameter correlation. Background Technology

[0002] Ultrafiltration membrane technology is a precision membrane separation technology driven by pressure difference. The pore size is concentrated in 2-100nm. It can efficiently remove pollutants such as suspended solids, colloids, bacteria, and macromolecular organic matter in water. Compared with the traditional coagulation-sedimentation-sand filtration process, it has significant advantages such as high separation efficiency, no phase change, small footprint, low energy consumption, and simple operation and maintenance. It has been widely used in municipal sewage upgrading and renovation, deep reuse of industrial wastewater, drinking water purification, and seawater desalination pretreatment, becoming one of the core supporting technologies for high-quality development in the water treatment industry.

[0003] Membrane fouling is the core bottleneck restricting the large-scale and stable application of ultrafiltration systems during long-term continuous operation. Membrane fouling refers to the process by which pollutants in the water to be treated are adsorbed and deposited on the membrane surface and become embedded in the membrane pores, gradually forming a filter cake layer and a gel layer. Based on reversibility, it can be divided into reversible fouling that can be removed by hydraulic cleaning, semi-reversible fouling that can be restored by chemical cleaning, and irreversible fouling that causes permanent blockage of the membrane pores. As the degree of membrane fouling intensifies, ultrafiltration systems will experience a continuous decline in membrane flux and a significant increase in transmembrane pressure. This not only directly leads to a substantial increase in operating energy consumption and the frequency of chemical cleaning, increasing reagent consumption and maintenance costs, but also accelerates the aging and degradation of membrane materials, significantly shortening the service life of membrane modules. In severe cases, it can even lead to substandard permeate water and system shutdown, posing a significant risk to the stable operation of the project.

[0004] Currently, the core prerequisite for membrane fouling control is the accurate measurement and determination of the fouling state. Existing technologies are mainly offline detection, which requires shutting down and disassembling the membrane module to complete the detection through equipment such as scanning electron microscope and atomic force microscope, or judging by a single membrane pressure difference and membrane flux decay. These technologies have the drawbacks of being complex to operate, having long detection cycles, strong lag, and being unable to achieve online continuous monitoring, making it difficult to meet the operation and maintenance needs of engineering sites. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for determining membrane fouling state based on multidimensional operating parameter correlation, used for monitoring and evaluating ultrafiltration membranes. The method for determining membrane fouling state based on multidimensional operating parameter correlation includes: S1. Obtain multi-dimensional input parameters of the ultrafiltration membrane system under its current operating status.

[0006] S2. Preprocess the obtained multidimensional input parameters to obtain a multidimensional parameter dataset.

[0007] S3. Input the multidimensional parameter dataset into the pre-defined optimal analysis model to obtain the current membrane fouling index PI.

[0008] S4. Compare the current membrane fouling index PI with the preset range of each fouling level to quantitatively determine the fouling status level of the ultrafiltration membrane, and determine the treatment measures based on the fouling status level.

[0009] Preferred steps: The preprocessing steps are as follows: (1) Calculation and filling of key parameters; (2) Definition and calculation of membrane fouling index; (3) Handling missing values; (4) Outlier handling.

[0010] Preferred: Membrane fouling index

[0011] Where: PI is the membrane fouling index; g is the fouling spreading distribution coefficient; J0 is the initial membrane flux; J is the membrane flux; p0 is the initial membrane pressure difference; p is the membrane pressure difference; V sp This refers to the water production per unit area of ​​the membrane.

[0012] Preferred: Pollution paving distribution coefficient Where A0 is the effective filtration area of ​​the ultrafiltration membrane; β is the resistance coefficient per unit thickness of the sludge; and Q is the total volume of liquid filtered at the current time point. α is the sludge volume concentration; γ is the time accumulation base; t is the filtration accumulation time.

[0013] Preferred: The optimal analysis model acquisition process includes: Data is extracted from the ultrafiltration membrane to be tested to obtain operating condition data.

[0014] The operating condition data is preprocessed to obtain the operating condition sample dataset.

[0015] Then, a neural network model is trained to obtain the optimal analysis model.

[0016] Preferred: Neural network model training includes: (1) Data set partitioning: The preprocessed working condition sample dataset is divided into a training set and a test set according to a preset ratio. The training set is used for model training, and the test set is used for model performance evaluation. (2) Neural network model construction: Construct a 3-layer neural network model; (3) Model training and optimization: The constructed 3-layer neural network model is trained using the training set; (4) Model adaptation and adjustment: Update the relevant parameters of the model according to the process parameters and sampling time of the specific application scenario.

[0017] Preferred operating data includes historical operating data and multi-condition experimental data.

[0018] Preferred operating data and multidimensional input parameters include: time since last cleaning, COD value, ammonia nitrogen ratio, sludge concentration, influent flow rate, temperature, influent pressure, permeate flow rate, and concentrate pressure.

[0019] The preferred ratio is 7:3 between the training set and the test set.

[0020] Preferred: The construction of a 3-layer neural network model includes: an input layer with 6 nodes, selecting the 6 core parameters that have the most significant impact on membrane fouling as input variables; an output layer with 1 node, the output variable being the membrane fouling index PI, used to directly characterize the membrane fouling state; and a hidden layer with 7 nodes, used to realize the nonlinear mapping between input and output parameters; except for the input layer, each node in each layer contains one nonlinear Sigmoid transform.

[0021] The preferred parameters are: COD value, sludge concentration, influent flow rate, temperature, influent pressure, and concentrate pressure.

[0022] Preferred: The pollution status levels include four levels: Level I (clean), Level II (lightly polluted), Level III (moderately polluted), and Level IV (severely polluted).

[0023] Preferred: Level I cleanliness corresponds to a pollution level range of 0 ≤ PI ≤ 0.1.

[0024] Preferred: Level II light pollution corresponds to a pollution level range of 0.1. <PI≤0.3。

[0025] Preferred: Level III moderate pollution corresponds to a pollution level range of 0.3. <PI≤0.6。

[0026] Preferred: Level IV severe pollution corresponds to a pollution level range PI > 0.6.

[0027] Preferred measures for Level I cleanliness include: implementing standard daily operating procedures, conducting routine backwashing according to preset cycles without additional treatment; strictly controlling influent water quality (COD, sludge concentration), influent flow rate, and pressure within the design range to avoid overload operation; maintaining real-time measurement of the current membrane fouling index (PI) in the neural network model, recording operating parameters every 2 hours, tracking the changing trend of the current PI value, and initiating preventive maintenance in advance if a continuous upward trend is observed; model updates: after each chemical cleaning, recalibrating the initial membrane pressure difference P0 and initial membrane flux J0, updating model parameters, and ensuring the accuracy of the current PI measurement.

[0028] Preferred treatment measures for Level II mild contamination include: Backwash optimization: Increase the conventional backwash flow rate to 1.1-1.2 times the design backwash flow rate, extend the backwash time from 30s to 60-90s, shorten the backwash cycle from 30min to 15-20min, and perform 3-5 consecutive cycles; Combined air-water cleaning: For hollow fiber membranes, activate air washing, controlling the air washing intensity at 15-20L / (m³). 2 s), air-water mixed cleaning time 60~120s, using the shear force of air bubbles to peel off the loose filter cake layer on the membrane surface; drain soaking: after completing the air-water cleaning, soak the membrane module with ultrafiltration permeate for 10~15min, then drain to further flush away the detached contaminants. Maintenance chemical enhanced backwash (standby): add a low concentration of reagent (50~100mg / L sodium hypochlorite, or 0.1%~0.2% citric acid) to the backwash water, soak for 5~10min, then backwash to remove weakly adsorbed organic matter and prevent further contamination.

[0029] Preferred treatment measures for Level III moderate pollution include: Preparation before cleaning: Stop the production water operation, rinse the membrane module with ultrafiltration permeate for 3-5 minutes, and drain the raw water from the module to prevent raw water pollutants from consuming the cleaning agent. Phased online chemical cleaning (suitable for mainstream pollutants in municipal wastewater): Phase 1: Alkaline washing and sterilization to remove organic matter and microbial slime. Prepare a cleaning solution of 0.5%~1.0% sodium hydroxide + 0.05%~0.1% sodium hypochlorite, circulate for 30 minutes, soak for 60-90 minutes, recirculate for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral. Phase 2: Acid washing and descaling to remove inorganic salt scale and metal oxides. Prepare a cleaning solution of 0.5%~1.0% citric acid (or 0.2% hydrochloric acid), circulate for 30 minutes, soak for 30-60 minutes, recirculate for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral. Root cause control: Investigate the causes of fluctuations in influent water quality, adjust the front-end pretreatment process (such as coagulation sedimentation and filtration), strictly control the influent COD and sludge concentration, and prevent the pollution from escalating rapidly again.

[0030] Preferred: Level IV severe fouling includes: Enhanced online chemical cleaning (preferred attempt): Increase reagent concentration: increase sodium hydroxide concentration for alkaline washing to 1.0%~2.0%, sodium hypochlorite concentration to 0.1%~0.2%; increase citric acid concentration for acid washing to 1.0%~2.0%; Extend soaking time: extend alkaline washing soaking time to 2~4 hours, acid washing soaking time to 1~2 hours, and perform two cycles of alternating circulation and soaking; Targeted reagent adaptation: if special contaminants (such as grease, recalcitrant organic matter) are present, add 0.05%~0.1% nonionic surfactant to improve cleaning effect. Offline chemical cleaning (performed when enhanced cleaning is ineffective): If the current membrane fouling index (PI) is still higher than 0.15 after two enhanced online cleanings, it indicates that severe blockage has formed in the membrane pores. The membrane module needs to be disassembled and sent to a professional cleaning workshop for offline processes such as segmented soaking, ultrasonic cleaning, and high-pressure pulse cleaning to deeply remove contaminants from the membrane pores; after offline cleaning, test the membrane flux recovery rate. If the recovery rate is lower than 80%, it indicates that the membrane has suffered irreversible damage. If, after offline cleaning, the membrane flux recovery rate remains below 70%, the permeate quality continues to fail to meet standards, and the operating energy consumption exceeds the design value by more than 50%, it indicates that the membrane modules have reached the end of their service life and require batch replacement. After replacing the membrane modules, the initial membrane pressure difference P0 and initial membrane flux J0 should be recalibrated, and the neural network model should be retrained to adapt to the operating characteristics of the new membrane. System rectification: A comprehensive investigation of defects in system design, operating parameters, and front-end pretreatment should be conducted, and the operating process should be optimized to avoid the recurrence of severe pollution.

[0031] The technical effects and advantages of this invention are as follows: 1. Solved the problem of limited historical operating condition data: By conducting multi-condition experiments, experimental data under different loads and water quality conditions were supplemented, and a comprehensive dataset was constructed by combining historical operating condition data, which effectively improved the generalization ability and measurement accuracy of the model and avoided measurement bias caused by modeling with single data. 2. Improved data preprocessing process: To address the issue of missing on-site measurement devices, key parameters (outlet pressure, membrane pressure difference) were obtained through principle deduction; combined with the ultrafiltration water quality change pattern, the regression imputation method was used to process missing data, and the method of "treating outliers as missing values" was used to process abnormal data, so as to retain the original data characteristics to the maximum extent, ensure the integrity and accuracy of the data, and provide reliable support for model training; 3. The model is reasonably designed and highly adaptable: The constructed 3-layer neural network model, combined with the Sigmoid nonlinear transformation, is adapted to the quantification range of the membrane fouling index. The gradient descent method is used to optimize the model parameters and improve the model's measurement accuracy. At the same time, the model can update parameters according to specific processes and sampling times to adapt to the operating requirements of different ultrafiltration systems. 4. Simple operation and high practicality: The entire measurement method does not require shutdown. Based on the multi-dimensional operating parameters that can be collected on site, the model realizes the accurate measurement of membrane fouling status. It is simple to operate and highly efficient, and can provide timely and reliable basis for the operation and maintenance of ultrafiltration system, reduce operating energy consumption and maintenance costs, and extend the service life of ultrafiltration membrane.

[0032] 5. By coupling the membrane fouling index with the relative change in membrane pressure, the relative decline in membrane flux, and the distribution state, the degree of membrane fouling can be comprehensively reflected in three dimensions, avoiding misjudgment based on a single parameter; the use of dimensionless relative values ​​makes it unaffected by differences in operating conditions and equipment, resulting in more stable and reliable characterization; by indirectly calculating key pressure parameters, it avoids the limitation that concentrate pressure cannot directly and comprehensively characterize fouling; by outputting a unique quantitative value, it is compatible with neural network model measurement and intelligent operation and maintenance, enabling accurate fouling classification and on-demand cleaning, significantly improving the economic efficiency and intelligence level of ultrafiltration system operation. Attached Figure Description

[0033] Figure 1 This is a schematic flowchart of a membrane fouling state determination method based on multidimensional operating parameter correlation proposed in this invention.

[0034] Figure 2 This is a schematic diagram of the pretreatment process in a membrane fouling state determination method based on multidimensional operating parameter correlation proposed in this invention.

[0035] Figure 3 This is a schematic diagram of the process for obtaining the optimal analytical model in a membrane fouling state determination method based on multidimensional operating parameter correlation proposed in this invention.

[0036] Figure 4 This is a schematic diagram of the neural network model training process in a membrane fouling state determination method based on multidimensional operating parameter correlation proposed in this invention. Detailed Implementation

[0037] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and should not be construed as limiting the invention. Rather, embodiments of the invention include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

[0038] Example 1 refer to Figure 1 This embodiment proposes a membrane fouling state determination method based on multidimensional operating parameter correlation for monitoring and evaluating ultrafiltration membranes. The membrane fouling state determination method based on multidimensional operating parameter correlation includes: S1. Obtain multi-dimensional input parameters of the ultrafiltration membrane system under its current operating status (consistent with the input parameter types used during model training), including but not limited to the time since the last cleaning, COD value, ammonia nitrogen ratio, sludge concentration, influent flow rate, temperature, influent pressure, permeate flow rate, and concentrate pressure; save these parameters as historical operating data. For example, obtain the current operating parameters of the ultrafiltration system: COD value = 45 mg / L, sludge concentration = 3200 mg / L, influent flow rate = 420 m³ / L. 3 / h, temperature = 22℃, inlet water pressure = 0.16MPa, concentrate pressure = 0.09MPa, etc.

[0039] S2. Preprocess the obtained multidimensional input parameters to obtain a multidimensional parameter dataset. This addresses issues such as missing key parameters, missing data, and data anomalies, ensuring data integrity and accuracy. (Reference) Figure 2 The preprocessing steps for multidimensional input parameters can be as follows: 1. Key Parameter Calculation and Filling: For example, considering that no outlet water pressure measuring device is installed on site, and the outlet water pressure cannot be directly obtained, the outlet water pressure of the ultrafiltration membrane is calculated based on the operating principle of the ultrafiltration membrane. The specific formula is: Outlet water pressure = Inlet water pressure - Concentrate water pressure; Based on this, the ultrafiltration membrane pressure difference is calculated. The specific formula is: Membrane pressure difference = Inlet water pressure - Outlet water pressure. In a certain set of data, the inlet water pressure is 0.15 MPa, and the concentrate water pressure is 0.08 MPa. Then, the outlet water pressure = 0.15 - 0.08 = 0.07 MPa, and the membrane pressure difference = 0.15 - 0.07 = 0.08 MPa.

[0040] 2. Definition and Calculation of Membrane Fouling Index: The optimal operating condition of the ultrafiltration membrane from the experimental process or historical data is selected as the initial state. The initial membrane pressure difference and initial membrane flux under this condition are recorded. Based on the ultrafiltration membrane fouling mechanism, the membrane fouling index formula is derived. The membrane fouling index is used to quantitatively characterize the degree of fouling of the ultrafiltration membrane. The formula is as follows:

[0041] In the formula: PI is the membrane fouling index, in meters. 2 / L; g is the pollution spread distribution coefficient, dimensionless; J0 is the initial membrane flux, L / (m²). 2 ·h); J is the membrane flux, L / (m 2 ·h); p0 is the initial membrane pressure difference, Pa; p is the membrane pressure difference, Pa; V sp The water production rate per unit area of ​​the membrane, in L / m². 2A higher PI value indicates a more severe degree of membrane fouling, while a PI value of 0 indicates that the membrane is in an unfouled initial state. For example, the initial state was selected as day 5 of the experiment, with the circulating pump at 70% load, optimal effluent quality, and most stable membrane flux. The initial membrane pressure difference ΔP0 = 0.05 MPa and the initial membrane flux J0 = 80 L / (m²) were recorded. 2 ·h), calculate the membrane fouling index PI for each sample group according to the formula. The specific calculation process will not be described here.

[0042] The contamination distribution coefficient *g* characterizes the distribution of contaminants during ultrafiltration. If the contamination distribution is uniform, the coefficient is 1. If the distribution is uneven, it needs to be calculated. We can divide the ultrafiltration membrane into regions and then analyze and calculate the values ​​at each location. However, this method requires testing each region individually, which is difficult to implement in actual production. The specific calculation process can be described using the contamination distribution coefficient... Where A0 is the effective filtration area of ​​the ultrafiltration membrane, which can be calculated based on the area of ​​the ultrafiltration membrane that is contaminated within a preset time period (usually 24 hours). β is the sludge unit thickness impedance coefficient, which can be understood as the ratio of the membrane pressure difference increased by a unit thickness of sludge to the initial membrane pressure difference. Its value can be obtained experimentally, although other values ​​are not excluded, which will not be elaborated here. Q is the total liquid volume filtered at the current time point, and its value can be calculated by multiplying the filtration time and the flow rate, which will not be elaborated here. α represents the sludge volume concentration. γ is the time accumulation base, with a value of (0,1), which can be obtained experimentally. Its value needs to be determined based on parameters such as the filtration accumulation time unit and sludge concentration; details will not be elaborated here. γ is the adjustment coefficient, with a value of (0,10), which needs to be obtained experimentally or empirically; details will not be elaborated here. t is the filtration accumulation time, which can be in days; details will not be elaborated here. This formula is used to calculate the contamination spreading distribution coefficient, which characterizes the sludge distribution, reflecting the initial near-uniform distribution of sludge in the ultrafiltration membrane, the emergence of unevenness over time, and then a tendency towards uniformity as sludge accumulates. Of course, other calculation methods are not excluded. For example, for the same filtration system and filter element, a contamination spreading distribution coefficient distribution curve can be constructed based on time, and then the value can be obtained using the filtration accumulation time; details will not be elaborated here.

[0043] 3. Handling Missing Values: Due to objective factors such as data acquisition interruptions, human error, and instrument malfunctions during the experiment, data loss is likely to occur. Based on the on-site experimental analysis, the changes in ultrafiltration influent water quality have certain regularities. Therefore, the regression imputation method is used to handle missing data. By establishing a regression model, the missing data is inferred based on the existing complete data, so as to restore the original data characteristics as accurately as possible and avoid the reduction in data volume and data deviation caused by deleting missing data.

[0044] 4. Outlier Handling: For data anomalies caused by factors such as instrument malfunction, system signal acquisition delay, and network packet loss, the method of "treating outliers as missing values" is adopted. Subsequently, the regression imputation method mentioned above is used to impute the missing values. This avoids the data waste caused by directly deleting outliers and also avoids the interference of outliers on model training, ensuring the quality of preprocessed data. The preprocessed data is then integrated and unified to form a multidimensional parameter dataset.

[0045] S3. Input the multidimensional parameter dataset into the pre-defined optimal analysis model to obtain the current membrane fouling index PI. (Reference) Figure 3 The optimal analytical model construction method includes: extracting data from the ultrafiltration membrane to be tested to obtain operating condition data. Operating condition data can include historical operating condition data and multi-condition experimental data. Historical operating condition data is collected in real-time during the long-term operation of the ultrafiltration membrane system. Since the on-site ultrafiltration circulation pump often operates at full load, the historical operating condition data has a single operating environment. Therefore, it is necessary to design and conduct multi-condition experiments for a preset time. Multi-condition experiments simulate different operating conditions by adjusting the ultrafiltration circulation pump load, influent parameters, etc. Sampling and parameter collection are performed at fixed time intervals. The collected parameters are consistent with those in the historical operating condition data, namely, the time since the last cleaning, COD value, ammonia nitrogen ratio, sludge concentration, influent flow rate, temperature, influent pressure, permeate flow rate, and concentrate pressure. These serve as supplementary samples for model training, ensuring that the samples cover the parameter range under different operating loads and water quality conditions. Of course, using only historical operating data is not excluded. However, modeling solely based on on-site historical operating data suffers from limitations due to the long-term full-load operation of the ultrafiltration circulating pump, resulting in limited data coverage, poor model generalization ability, low measurement accuracy, and difficulty in adapting to varying operating conditions. Combining historical operating data with multi-condition experimental data significantly expands data diversity, strengthens the correlation between parameters, and improves model robustness and measurement accuracy, thereby achieving a more accurate and stable measurement and evaluation of ultrafiltration membrane fouling status. For example, collecting historical operating data from the ultrafiltration system of a municipal wastewater treatment plant over the past year, including parameters such as: time since last cleaning (d), COD value (mg / L), ammonia nitrogen ratio (NH3-N / COD), sludge concentration (MLSS, mg / L), and influent flow rate (m³ / kg / L). 3 / h), temperature (°C), inlet water pressure (MPa), product water flow rate (m³ / h) 3 The data included parameters such as influent flow rate and concentrate pressure (MPa). Statistical analysis showed that due to the long-term full-load operation of the ultrafiltration circulating pump, the historical data for parameters such as influent flow rate and concentrate pressure fluctuated within a range of less than 5%, resulting in relatively homogeneous data. A 30-day multi-condition experiment was conducted, adjusting the ultrafiltration circulating pump load (50%, 60%, 70%, 80%, 90%, 100%). Samples were taken at fixed 2-hour intervals daily, collecting nine parameters consistent with historical operating condition data, resulting in 360 sets of experimental data. Operating condition samples under different loads were supplemented and merged with the historical operating condition data (720 sets in total) to form a complete sample dataset of 1080 sets. The operating condition data underwent the preprocessing operations described in S2 above (key parameter estimation, outlier handling, and missing value imputation) to obtain the operating condition sample dataset. For example, membrane pressure differential calculation and membrane fouling index calculation were performed. Upon inspection, 42 sets of data (mainly missing COD values ​​and sludge concentrations) were found among the 1080 samples. Since municipal wastewater influent quality exhibits clear temporal regularities (e.g., stable fluctuations during the day and night), a linear regression imputation method was used. Based on complete data from adjacent sampling points within the same time interval, a linear regression model was established to estimate the missing data and complete the missing value imputation. For instance, the 3σ criterion was used to identify anomalous data, identifying 18 sets of anomalous data (mainly abnormal influent pressure and concentrate pressure). These 18 sets of anomalous data were considered missing values ​​and imputed using the aforementioned linear regression method, completing the data anomaly processing. Finally, a complete and accurate dataset of 1080 operating condition samples was obtained.

[0046] Based on a dataset of working condition samples, a neural network model is trained, and the optimal analysis model is obtained by optimizing the model parameters. (Reference) Figure 4 The specific process is divided into three parts. 1. Dataset Partitioning: The preprocessed working condition sample dataset is divided into a training set and a test set according to a preset ratio. The training set is used for model training, and the test set is used for model performance evaluation. Preferably, the ratio of the training set to the test set is 7:3. The preprocessed 1080 datasets are divided into the training set and the test set in a 7:3 ratio, with 756 sets in the training set and 324 sets in the test set.

[0047] 2. Neural Network Model Construction: A three-layer neural network model is constructed, with the following structure: The input layer has 6 nodes, selecting the 6 core parameters that have the most significant impact on membrane fouling as input variables (which can be selected from preprocessed multidimensional parameters, such as COD value, sludge concentration, influent flow rate, temperature, influent pressure, and concentrate pressure); the output layer has 1 node, and the output variable is the membrane fouling index PI, which is used to directly characterize the membrane fouling state; the hidden layer has 7 nodes, used to realize the nonlinear mapping between input and output parameters; except for the input layer, each node contains a nonlinear sigmoid transformation, which is monotonically increasing and can map any real number to the range of 0 to 1, adapting to the quantization range of the membrane fouling index PI and improving the stability of the model output.

[0048] 3. Model Training and Optimization: The constructed 3-layer neural network model is trained using a training set. Gradient descent is employed during training to minimize the error between the model's measured values ​​and actual values. Simultaneously, the performance of the trained model is evaluated using a test set, calculating metrics such as measurement error and accuracy. The "training-evaluation" step is repeated, adjusting model parameters (such as learning rate, number of iterations, and hidden layer node weights) until the model's measurement accuracy reaches the preset requirements, thus obtaining the optimal analysis model. For example, using gradient descent for model training, a learning rate of 0.01 and 1000 iterations are set, with mean squared error (MSE) used as the loss function to minimize the error between measured and actual PI values. During training, the model's performance is evaluated using a test set every 100 iterations, and the measurement accuracy is calculated. The learning rate and number of iterations are repeatedly adjusted. When the number of iterations reaches 800, the model's measurement accuracy on the test set reaches 96.3%, and the mean squared error (MSE) = 0.002, meeting the preset accuracy requirements. Training is then stopped, and the optimal analysis model is obtained.

[0049] 4. Model Adaptation and Adjustment: Since different ultrafiltration systems have varying specific processes and sampling time intervals, their operating characteristics will also differ. Therefore, the optimal analytical model trained needs to be updated with relevant parameters based on the process parameters and sampling time of the specific application scenario to ensure the model can adapt to changes in actual operating conditions and maintain high measurement accuracy. For example, based on the specific process of the ultrafiltration system in this municipal wastewater treatment plant (treatment capacity 10000m³),... 3The model parameters are updated according to the sampling time interval (2 hours) and the ultrafiltration membrane model (PVDF hollow fiber membrane) to ensure that the model is adapted to the operating conditions of the system. The preprocessed current parameters are input into the trained optimal analysis model. After the model performs calculations, it produces a unique output value, namely the current membrane fouling index PI. The current parameters are preprocessed (estimated effluent pressure = 0.07MPa, membrane pressure difference = 0.09MPa, no missing or abnormal data); the preprocessed current multidimensional input parameters are input into the optimal analysis model. After the model performs calculations, it outputs the current membrane fouling index PI = 0.32.

[0050] S4. Compare the current membrane fouling index (PI) with the preset ranges for each fouling level to quantitatively determine the fouling status of the ultrafiltration membrane, and determine the treatment measures based on the fouling status. Each fouling status and corresponding refined treatment measures may include the following: Level I cleanliness (0≤PI≤0.1) Core Objectives: Maintain membrane cleanliness, slow down fouling development, and ensure effective daily operation and maintenance management. Daily Operation and Maintenance Procedures: Execute standard daily operating procedures, conduct routine backwashing according to preset cycles (no additional treatment required); strictly control influent water quality (COD, sludge concentration), influent flow rate, and pressure within design limits to avoid overload operation; maintain real-time measurement of the current membrane fouling index (PI) in the neural network model, record operating parameters every 2 hours, track the trend of the current PI value, and initiate preventative maintenance in advance if a continuous upward trend is observed; Model Updates: After each chemical cleaning, recalibrate the initial membrane pressure difference P0 and initial membrane flux J0, update model parameters, and ensure the accuracy of the current PI measurement.

[0051] Level II, Light Pollution (0.1) <PI≤0.3) Core objective: To completely remove reversible contaminants through hydraulic cleaning, prevent further contamination, restore the membrane's initial performance, and avoid membrane damage caused by chemical cleaning.

[0052] Backwash optimization: Increase the conventional backwash flow rate to 1.1~1.2 times the design backwash flow rate, extend the backwash time from 30s to 60~90s, shorten the backwash cycle from 30min to 15~20min, and perform 3~5 consecutive cycles; Air-water combined cleaning: For hollow fiber membranes, start air washing, and control the air washing intensity at 15~20L / (m³). 2 s), air-water mixed cleaning time 60~120s, using the shear force of air bubbles to peel off the loose filter cake layer on the membrane surface; drain soaking: after completing the air-water cleaning, soak the membrane module with ultrafiltration permeate for 10~15min, then drain to further flush away the detached contaminants. Maintenance chemical enhanced backwash (standby): add a low concentration of reagent (50~100mg / L sodium hypochlorite, or 0.1%~0.2% citric acid) to the backwash water, soak for 5~10min, then backwash to remove weakly adsorbed organic matter and prevent further contamination.

[0053] Level III, Moderate Pollution (0.3) <PI≤0.6) Core Objective: To remove semi-reversible contaminants through online chemical cleaning, fully restore membrane performance, and prevent the development of irreversible fouling. This is the standard trigger point for chemical cleaning in the industry. Preparation before cleaning: Stop the permeate operation and rinse the membrane module with ultrafiltration permeate for 3-5 minutes. Drain the raw water from the module to prevent raw water contaminants from consuming the cleaning agent. Phased online chemical cleaning (suitable for mainstream pollutants in municipal wastewater): Phase 1: Alkaline washing and sterilization to remove organic matter and microbial slime. Prepare a cleaning solution of 0.5%~1.0% sodium hydroxide + 0.05%~0.1% sodium hypochlorite, circulate for 30 minutes, soak for 60-90 minutes, recirculate for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral. Phase 2: Acid washing and descaling to remove inorganic salt scale and metal oxides. Prepare a cleaning solution of 0.5%~1.0% citric acid (or 0.2% hydrochloric acid), circulate for 30 minutes, soak for 30~60 minutes, circulate again for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral. Root cause control: Investigate the causes of influent water quality fluctuations, adjust upstream pretreatment processes (such as coagulation sedimentation, filtration), strictly control influent COD and sludge concentration, and prevent rapid escalation of pollution.

[0054] Level IV Severe Pollution (PI>0.6) Core Objective: To salvage membrane performance through enhanced cleaning, assess membrane module lifespan, and prevent system failure; this is an emergency treatment stage for irreversible contamination. Enhanced Online Chemical Cleaning (Preferred): Increase reagent concentration: Increase sodium hydroxide concentration for alkaline cleaning to 1.0%~2.0%, sodium hypochlorite concentration to 0.1%~0.2%; increase citric acid concentration for acid cleaning to 1.0%~2.0%; Extend soaking time: Extend alkaline cleaning soaking time to 2~4 hours, acid cleaning soaking time to 1~2 hours, and perform two cycles of alternating circulation and soaking; Targeted reagent selection: If special contaminants (such as grease, recalcitrant organic matter) are present, add 0.05%~0.1% nonionic surfactant to improve cleaning effectiveness. Offline chemical cleaning (performed when enhanced cleaning is ineffective): If the membrane fouling index (PI) remains above 0.15 after two enhanced online cleaning cycles, it indicates severe blockage within the membrane pores. The membrane module must be disassembled and sent to a professional cleaning workshop for offline processes such as segmented immersion, ultrasonic cleaning, and high-pressure pulse cleaning to deeply remove contaminants from the membrane pores. After offline cleaning, the membrane flux recovery rate is tested. If the recovery rate is below 80%, irreversible damage to the membrane has occurred. If the membrane flux recovery rate remains below 70% after offline cleaning, the permeate water quality continues to fail to meet standards, and operating energy consumption exceeds the design value by more than 50%, the membrane module has reached the end of its service life and requires batch replacement. After replacing the membrane module, the initial membrane pressure differential (P0) and initial membrane flux (J0) are recalibrated, and the neural network model is retrained to adapt to the operating characteristics of the new membrane. System rectification: A comprehensive review of system design, operating parameters, and front-end pretreatment defects is conducted, and the operating process is optimized to prevent the recurrence of severe fouling. This is just a simple example and not universally applicable; other situations will not be elaborated upon here.

[0055] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0056] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for determining membrane fouling state based on multidimensional operating parameter correlation, characterized in that, The membrane fouling state determination method based on multidimensional operating parameter correlation includes: S1. Obtain multi-dimensional input parameters of the ultrafiltration membrane system under its current operating status; S2. Preprocess the obtained multidimensional input parameters to obtain a multidimensional parameter dataset; Preprocessing steps include: (1) Calculation and filling of key parameters; (2) Definition and calculation of membrane fouling index; (3) Handling missing values; (4) Outlier handling; S3. Input the multidimensional parameter dataset into the pre-defined optimal analysis model to obtain the current membrane fouling index PI; S4. Compare the current membrane fouling index PI with the preset range of each fouling level to quantitatively determine the fouling status level of the ultrafiltration membrane, and determine the treatment measures based on the fouling status level. Membrane fouling index Where: PI is the membrane fouling index; g is the fouling spreading distribution coefficient; J0 is the initial membrane flux; J is the membrane flux; p0 is the initial membrane pressure difference; p is the membrane pressure difference; V sp This refers to the water production per unit area of ​​the membrane.

2. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 1, characterized in that, The process for obtaining the optimal analysis model includes: Data is extracted from the ultrafiltration membrane to be tested to obtain operating condition data; Preprocess the operating condition data to obtain the operating condition sample dataset; Then, a neural network model is trained to obtain the optimal analysis model.

3. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 2, characterized in that, Neural network model training includes: (1) Data set partitioning: The preprocessed working condition sample dataset is divided into a training set and a test set according to a preset ratio. The training set is used for model training, and the test set is used for model performance evaluation. (2) Neural network model construction: Construct a 3-layer neural network model; (3) Model training and optimization: The constructed 3-layer neural network model is trained using the training set; (4) Model adaptation and adjustment: Update the relevant parameters of the model according to the process parameters and sampling time of the specific application scenario.

4. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 2, characterized in that, The operating condition data includes historical operating condition data and multi-condition experimental data.

5. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 1, characterized in that, The pollution status levels include four levels: Level I (clean), Level II (lightly polluted), Level III (moderately polluted), and Level IV (severely polluted).

6. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 5, characterized in that, Level I cleanliness measures include: implementing standard daily operating procedures, conducting routine backwashing according to preset cycles without additional treatment; strictly controlling influent water quality, influent flow rate, and pressure within the design range to avoid overload operation; maintaining real-time measurement of the current membrane fouling index (PI) in the neural network model, recording operating parameters every 2 hours, tracking the changing trend of the current PI value, and initiating preventive maintenance in advance if a continuous upward trend is observed; model updates: after each chemical cleaning, recalibrating the initial membrane pressure difference P0 and initial membrane flux J0, updating model parameters, and ensuring the accuracy of the current PI measurement.

7. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 5, characterized in that, Level II mild contamination treatment measures include: Backwash optimization: increasing the conventional backwash flow rate to 1.1-1.2 times the design backwash flow rate, extending the backwash time from 30 seconds to 60-90 seconds, shortening the backwash cycle from 30 minutes to 15-20 minutes, and performing 3-5 consecutive cycles; Combined air-water cleaning: for hollow fiber membranes, initiating air washing, with the air washing intensity controlled at 15-20 L / m³. 2 The air-water mixed cleaning time is 60-120 seconds, using the shear force of air bubbles to peel off the loose filter cake layer on the membrane surface; evacuation soaking: after completing the air-water cleaning, the membrane module is soaked in ultrafiltration permeate for 10-15 minutes, and then evacuated to further flush away the detached pollutants; maintenance chemical enhanced backwashing: a low concentration of chemical agent is added to the backwash water, soaked for 5-10 minutes, and then backwashed to remove weakly adsorbed organic matter and prevent further contamination.

8. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 5, characterized in that, The treatment measures for Level III moderate pollution include: Preparation before cleaning: Stop the production water operation, rinse the membrane module with ultrafiltration permeate for 3-5 minutes, and drain the raw water from the module to prevent raw water contaminants from consuming cleaning agents; Staged online chemical cleaning: Stage 1: Alkaline washing and sterilization to remove organic matter and microbial slime. Prepare a cleaning solution of 0.5%-1.0% sodium hydroxide + 0.05%-0.1% sodium hypochlorite, circulate for 30 minutes, soak for 60-90 minutes, recirculate for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral; Stage 2: Acid washing and descaling to remove inorganic salt scale and metal oxides. Prepare a cleaning solution of 0.5%-1.0% citric acid, circulate for 30 minutes, soak for 30-60 minutes, recirculate for 30 minutes, and rinse with ultrafiltration permeate until the effluent pH is neutral; Source control: Investigate the causes of influent water quality fluctuations, adjust the front-end pretreatment process, strictly control influent COD and sludge concentration, and prevent rapid escalation of pollution.

9. The method for determining membrane fouling state based on multidimensional operating parameter correlation according to claim 5, characterized in that, Level IV severe contamination includes: Enhanced online chemical cleaning: Increased reagent concentration: Alkaline washing sodium hydroxide concentration increased to 1.0%~2.0%, sodium hypochlorite concentration increased to 0.1%~0.2%; Acid washing citric acid concentration increased to 1.0%~2.0%; Extended soaking time: Alkaline washing soaking time extended to 2~4 hours, acid washing soaking time extended to 1~2 hours, cycle-soaking alternating twice; Targeted reagent adaptation: If special contaminants are present, add 0.05%~0.1% nonionic surfactant to improve cleaning effect; Offline chemical cleaning: If the current membrane fouling index (PI) is still higher than 0.15 after two enhanced online cleanings, the membrane needs to be... The membrane modules are disassembled and sent to a professional cleaning workshop where offline processes such as segmented soaking, ultrasonic cleaning, and high-pressure pulse cleaning are used to deeply remove contaminants from the membrane pores. After offline cleaning, the membrane flux recovery rate is tested. If the membrane flux recovery rate is still below 70% after offline cleaning, the permeate water quality continues to fail to meet standards, and the operating energy consumption exceeds the design value by more than 50%, it indicates that the membrane modules have reached the end of their service life and need to be replaced in batches. After replacing the membrane modules, the initial membrane pressure difference P0 and initial membrane flux J0 are recalibrated, and the neural network model is retrained to adapt to the operating characteristics of the new membrane. System rectification: A comprehensive investigation of defects in system design, operating parameters, and front-end pretreatment is conducted, and the operating process is optimized.

Images