Membrane processing performance prediction system and prediction method based on machine learning

Abstract

The invention discloses a membrane processing performance prediction system based on machine learning in the technical field of membrane processing, and the system comprises a model characteristic detection end, a pollutant property detection and calculation end, an operation condition collection end, a cloud calculation learning end, a data output end, and a result feedback correction end. The method comprises the following steps: firstly, collecting membrane treatment conditions through membrane material characterization, membrane characteristic analysis, pollutant characteristic analysis, pollutant descriptor calculation and operation condition monitoring to obtain basic condition parameters; receiving data through a cloud computing learning end, and adopting a statistical method of machine learning to construct a model to calculate and predict a membrane removal rate, a water flux and a balance point between the removal rate and the water flux; and the cloud computing learning end is monitored in real time through the result feedback end, the predicted model and algorithm are continuously corrected, and finally the prediction result of the optimal membrane processing performance and the membrane processing condition are obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of membrane treatment, and in particular to a membrane treatment performance prediction system and prediction method. Background Art

[0002] Membrane separation technology is widely used in the treatment of drinking water and industrial wastewater, and is considered to be one of the most effective water treatment methods. Removal rate and water flux are commonly used parameters to measure membrane treatment performance. Among them, the increase in membrane water flux is usually at the expense of membrane removal rate, and vice versa. Therefore, membranes with higher pollutant removal efficiency often have lower water flux, which corresponds to higher energy consumption and economic costs. According to actual needs, designing the balance point between water flux and removal rate, that is, setting a suitable water flux target value, and optimizing the design and operating parameters of the membrane on this basis to maximize the removal rate is crucial to achieving more energy-saving and cost-effective membrane treatment. For example, in drinking water treatment, we may aim for a higher water flux, but at the same time ensure sufficient pollutant removal effect. In some wastewater reuse occasions, more attention may be paid to achieving a higher removal rate.

[0003] Membrane treatment of pollutants is a rather complex process, and membrane treatment performance is affected by membrane characteristics, pollutant characteristics and operating conditions. Optimizing multiple influencing factors is crucial to improving membrane treatment performance. Conventional experiments are difficult to fully consider the impact of multiple factors on membrane treatment performance indicators, and are accompanied by huge costs and resource consumption. Depending on actual needs, the balance point between removal rate and water flux will also vary, and the diversity of membrane treatment performance indicators further exacerbates the difficulty of experimental optimization. Therefore, an efficient tool is needed to improve membrane treatment performance.

[0004] In the past decade, machine learning has achieved rapid growth in various applications such as image classification and machine translation. From chemistry, materials science, biomedicine to quantum physics, machine learning is profoundly changing the research progress in many scientific fields. Machine learning, a data analysis method with low reliance on prior knowledge, shows great promise in solving complex data analysis problems due to its powerful fitting ability. This method can provide insights into the main factors affecting membrane treatment performance, reduce or replace related experiments, make up for the lack of experimental data, and reduce experimental costs.

[0005] At present, there are difficulties in optimizing the membrane treatment process, improving the pollutant removal rate, water flux, and finding a balance between the removal rate and water flux. The present invention, based on an improved membrane treatment performance prediction method, introduces machine learning, matches a basic experimental test device, automatically inputs parameters, and obtains membrane treatment performance data. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a membrane treatment performance prediction system and prediction system based on machine learning, which overcomes the time-consuming and labor-intensive problems of the existing testing methods. The experimental measurement device is combined with machine learning technology, and the prediction accuracy of membrane treatment performance is effectively improved through the rational use of artificial intelligence machines.

[0007] The object of the present invention is achieved by: a membrane treatment performance prediction system based on machine learning, characterized in that it includes: The model property detection end includes a membrane material characterization and membrane property analysis system, which is used to characterize and analyze the membrane contact angle, molecular weight cutoff, total charge, zeta potential, thermal conductivity and other properties of the membrane material; The pollutant property detection and calculation end includes a pollutant property analysis system and a pollutant descriptor calculation system, which are used to measure and analyze the molecular weight, compound size, compound charge, water solubility, octanol-water partition coefficient and other properties of the pollutant, and calculate the molecular structure descriptor of the pollutant; The operating condition acquisition end includes an operating condition monitoring system for measuring conditions such as pH, temperature, flow rate, osmotic pressure, operating time, cross flow velocity, ion concentration, natural organic matter concentration, and interaction force between membrane and pollutants during membrane treatment; The cloud computing learning end includes a cloud service computer cluster, which is used to receive the membrane characteristics, pollutant characteristics, and operating condition data from the model property detection end, the pollutant property detection and calculation end, and the operating condition collection end, and adopts the statistical method of machine learning to build a model to calculate and predict the membrane removal rate and water flux, and develop a model to design and predict the balance point between the removal rate and the water flux; The data output terminal includes a monitoring and result output system for monitoring the computing status of the cloud computing learning terminal and allocating computing cluster node resources according to the computing amount; The feedback correction end includes a big data system and an experimental test bench, which are used to compare the prediction results with the actual experimental results and continuously correct the prediction model and algorithm.

[0008] Furthermore, the membrane material characterization and membrane property analysis system at the membrane property detection end includes a contact angle meter, a permeation flux test system, an atomic force microscope, a zeta potential analyzer, and a thin film thermal conductivity test system, which can obtain the membrane contact angle, molecular weight cutoff, total charge, zeta potential, and thermal conductivity of the membrane material to be tested.

[0009] Furthermore, the pollutant property analysis system at the pollutant property detection and calculation end includes a pollutant molecular weight collection system, a transmission electron microscope, an X-ray photoelectron spectrometer, a UV-Vis spectrophotometer, and a high performance liquid chromatograph, which can accurately measure the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of the pollutant; the pollutant descriptor calculation system is alvaDesc software, which can obtain 5290 descriptors that quantify molecular structure feature information.

[0010] Furthermore, the operating condition monitoring system of the operating condition collection end includes a multi-parameter online water quality detector, a rotor flowmeter, an osmometer, an electromagnetic flowmeter, an ion chromatograph, a UV-visible spectrophotometer, and a surface force meter, which can monitor the pH, temperature, flow, osmotic pressure, operating time, cross-flow velocity, ion concentration, natural organic matter concentration, and the interaction force between membrane and pollutants during the membrane treatment process.

[0011] A prediction method of a membrane treatment performance prediction system based on machine learning, comprising the following steps: Step 1) First, use a contact angle meter, a permeation flux test system, an atomic force microscope, a zeta potential analyzer, and a thin film thermal conductivity test system to analyze and test the membrane material to be tested, and obtain membrane property data including membrane contact angle, molecular weight cutoff, total charge, zeta potential, and thermal conductivity; Step 2) The wastewater to be treated is then sent to the pollutant property detection and calculation end, and the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of the pollutants are measured using a molecular weight acquisition system, a transmission electron microscope, an X-ray photoelectron spectrometer, a UV-Vis spectrophotometer, and a high-performance liquid chromatograph; and 5290 molecular structure descriptors are calculated using alvaDesc software, integrated, and obtained pollutant characteristic data; Step 3) Afterwards, the pH, temperature, flow rate, osmotic pressure, operation time, cross flow velocity, ion concentration, natural organic matter concentration, and membrane-pollutant interaction force during the membrane treatment process are detected by a multi-parameter online water quality detector, a rotor flowmeter, an osmometer, an electromagnetic flowmeter, an ion chromatograph, a UV-visible spectrophotometer, and a surface force meter in the operation condition collection terminal to obtain membrane treatment operation data, and the membrane treatment operation data is recorded in real time; Step 4) The data obtained from the membrane characteristic detection end, the pollutant property detection and calculation end, and the operating condition collection end through the above operations are then transmitted to the cloud computing learning end. The statistical method of machine learning is adopted, the computing resources of the cloud service computer cluster are utilized, and according to the statistical model and parameters, the optimization algorithm is adopted to calculate and predict the membrane removal rate and water flux of the constructed model on the cloud service computer cluster, and a model is developed to design and predict the balance point between the removal rate and the water flux. The predicted data calculated by the cloud computing learning end is then displayed on the data output end. At the same time, the data output end can monitor the computing status of the cloud computing learning end and allocate computing cluster node resources according to the computing amount. The data output end then outputs the predicted membrane removal rate, water flux and removal rate. The result of the balance point between the removal rate and the water flux is fed back to the feedback correction end. On the one hand, the feedback correction end is connected to the big data system to obtain the removal rate, water flux and the balance point data between the removal rate and the water flux of each pollutant under different conditions. At the same time, the feedback correction section is also connected to the experimental test bench to obtain the removal rate, water flux and the balance point between the removal rate and the water flux obtained by the actual test; the feedback correction end compares the predicted result with the big data and the actual experimental results, and feeds back the result deviation to the cloud computing learning end. The cloud computing learning end continuously corrects the predicted model and algorithm to obtain the optimal membrane treatment removal rate, water flux and the balance point prediction model between the removal rate and the water flux, which is used for the prediction of membrane treatment removal rate, water flux and the balance point between the removal rate and the water flux.

[0012] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention does not require input parameters. All data required by the model are obtained through experimental testing, which can simulate the actual situation of membrane treatment and obtain accurate model parameters; (2) The present invention introduces membrane characteristics, pollutant characteristics and operating condition variable parameters, comprehensively considers the factors affecting membrane treatment performance in actual situations, and obtains more accurate membrane treatment performance; (3) The present invention uses artificial intelligence and machine learning to predict membrane treatment performance, which solves the time-consuming and labor-intensive defects of traditional single-factor experimental measurement and is more flexible and efficient; (4) The present invention can design the balance point between water flux and removal rate in membrane treatment according to actual needs, and efficiently and accurately obtain the membrane treatment conditions of the balance point; (5) The artificial intelligence and machine learning methods of the present invention compare the results with the experimental values ​​and correct the results, making the predictions more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0014] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] A membrane treatment performance prediction system based on machine learning, comprising: The model property detection end includes a membrane material characterization and membrane property analysis system, which is used to characterize and analyze the membrane contact angle, molecular weight cutoff, total charge, zeta potential, thermal conductivity and other properties of the membrane material; the membrane material characterization and membrane property analysis system at the membrane property detection end includes: contact angle meter, permeation flux test system, atomic force microscope, zeta potential analyzer, thin film thermal conductivity test system, which can obtain the membrane contact angle, molecular weight cutoff, total charge, zeta potential and thermal conductivity of the membrane material to be tested.

[0017] The pollutant property detection and calculation end includes a pollutant property analysis system and a pollutant descriptor calculation system, which are used to measure and analyze the molecular weight, compound size, compound charge, water solubility, octanol-water partition coefficient and other properties of pollutants, and calculate the molecular structure descriptors of pollutants. The pollutant property analysis system of the pollutant property detection and calculation end includes: a pollutant molecular weight acquisition system, a transmission electron microscope, an X-ray photoelectron spectrometer, a UV-Vis spectrophotometer, and a high-performance liquid chromatograph, which can accurately measure the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of pollutants; the pollutant descriptor calculation system is alvaDesc software, which can obtain 5290 descriptors of quantitative molecular structure feature information; The operating condition collection end includes an operating condition monitoring system, which is used to measure the pH, temperature, flow rate, osmotic pressure, operation time, cross flow velocity, ion concentration, natural organic matter concentration, membrane and pollutant interaction force and other conditions in the membrane treatment process. The operating condition monitoring system of the operating condition collection end includes a multi-parameter online water quality detector, a rotor flowmeter, an osmotic pressure meter, an electromagnetic flowmeter, an ion chromatograph, a UV-visible spectrophotometer, and a surface force meter, which can monitor the pH, temperature, flow rate, osmotic pressure, operation time, cross flow velocity, ion concentration, natural organic matter concentration, membrane and pollutant interaction force in the membrane treatment process; The cloud computing learning end includes a cloud service computer cluster, which is used to receive the membrane characteristics, pollutant characteristics, and operating condition data from the model property detection end, the pollutant property detection and calculation end, and the operating condition collection end. It uses the statistical method of machine learning to build a model to calculate and predict the membrane removal rate and water flux, and develops a model to design and predict the balance point between the removal rate and the water flux. The data output end includes a monitoring and result output system to monitor the computing status of the cloud computing learning end and allocate computing cluster node resources according to the computing volume; The feedback correction end includes a big data system and an experimental test bench, which is used to compare the predicted results with the actual experimental results and continuously correct the predicted model and algorithm.

[0018] like Figure 1 A prediction method of a membrane treatment performance prediction system based on machine learning is shown, comprising the following steps: Step 1) First, use a contact angle meter, permeation flux test system, atomic force microscope, zeta potential analyzer, and thin film thermal conductivity test system to analyze and detect the membrane material to be tested, and obtain membrane characteristic data including membrane contact angle, molecular weight cutoff, total charge, zeta potential, and thermal conductivity.

[0019] Step 2) The wastewater to be treated is then sent to the pollutant property detection and calculation end, and the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of the pollutants are measured using a molecular weight acquisition system, transmission electron microscope, X-ray photoelectron spectrometer, UV-Vis spectrophotometer, and high performance liquid chromatography; and 5290 molecular structure descriptors are calculated using alvaDesc software; and the pollutant characteristic data are integrated and obtained.

[0020] Step 3) After that, the pH, temperature, flow rate, osmotic pressure, operation time, cross flow velocity, ion concentration, natural organic matter concentration, and interaction force between membrane and pollutants in the membrane treatment process are detected by the multi-parameter online water quality detector, rotor flowmeter, osmometer, electromagnetic flowmeter, ion chromatograph, UV-visible spectrophotometer, and surface force meter in the operation condition collection end to obtain the membrane treatment operation data, and record the membrane treatment operation data in real time.

[0021] Step 4) The data obtained from the membrane characteristic detection end, the pollutant property detection and calculation end, and the operating condition collection end through the above operations are then transmitted to the cloud computing learning end, and statistical methods of machine learning, such as but not limited to neural networks, random forests, k-nearest neighbors, support vector machines, recurrent neural network algorithms, etc., are used. The computing resources of the cloud service computer cluster are utilized and based on statistical models and parameters, an optimization algorithm is used to calculate and predict the membrane removal rate and water flux of the constructed model on the cloud service computer cluster, and a model is developed to design and predict the balance point between the removal rate and the water flux; then the predicted data calculated by the cloud computing learning end is displayed on the data output end, and at the same time, the data output end can monitor the computing status of the cloud computing learning end and allocate computing cluster node resources according to the computing amount; then the data output end will The predicted membrane removal rate, water flux and the balance point between the removal rate and the water flux are fed back to the feedback correction end. The feedback correction end is connected to the big data system on the one hand to obtain the removal rate, water flux and the balance point data between the removal rate and the water flux of each pollutant under different conditions. At the same time, the feedback correction section is also connected to the experimental test bench to obtain the removal rate, water flux and the balance point between the removal rate and the water flux obtained from the actual test; the feedback correction end compares the predicted results with the big data and the actual experimental results, and feeds back the result deviation to the cloud computing learning end. The cloud computing learning end continuously corrects the predicted model and algorithm to obtain the optimal membrane treatment removal rate, water flux and the balance point prediction model between the removal rate and the water flux, which is used for the prediction of membrane treatment removal rate, water flux and the balance point between the removal rate and the water flux.

[0022] The membrane treatment performance prediction system combines the membrane characteristic detection end, the pollutant property detection and calculation end, the operating condition collection end, the data output end, and the feedback correction end to develop the optimal membrane treatment removal rate, water flux, and the balance point prediction model between the removal rate and water flux, which significantly improves the efficiency and accuracy of membrane treatment and meets different treatment effects. Multi-parameter detection ensures the comprehensiveness of data, intelligent prediction improves the treatment effect, and the self-optimization mechanism makes the model more adaptable. In addition, the efficient use of cloud computing resources and the flexible decision support system provide a scientific basis for the application of membrane treatment technology and have broad application potential.

[0023] The above embodiments are only used to help understand the method and core idea of ​​the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A membrane treatment performance prediction system based on machine learning, characterized in that: include: The model property detection end includes a membrane material characterization and membrane property analysis system, which is used to characterize and analyze the membrane contact angle, molecular weight cutoff, total charge, zeta potential, thermal conductivity and other properties of the membrane material; The pollutant property detection and calculation end includes a pollutant property analysis system and a pollutant descriptor calculation system, which are used to measure and analyze the molecular weight, compound size, compound charge, water solubility, octanol-water partition coefficient and other properties of the pollutant, and calculate the molecular structure descriptor of the pollutant; The operating condition acquisition end includes an operating condition monitoring system for measuring conditions such as pH, temperature, flow rate, osmotic pressure, operating time, cross flow velocity, ion concentration, natural organic matter concentration, and interaction force between membrane and pollutants during membrane treatment; The cloud computing learning end includes a cloud service computer cluster, which is used to receive the membrane characteristics, pollutant characteristics, and operating condition data from the model property detection end, the pollutant property detection and calculation end, and the operating condition collection end, and adopts the statistical method of machine learning to build a model to calculate and predict the membrane removal rate and water flux, and develop a model to design and predict the balance point between the removal rate and the water flux; The data output terminal includes a monitoring and result output system for monitoring the computing status of the cloud computing learning terminal and allocating computing cluster node resources according to the computing amount; The feedback correction end includes a big data system and an experimental test bench, which are used to compare the prediction results with the actual experimental results and continuously correct the prediction model and algorithm.

2. The membrane treatment performance prediction system based on machine learning according to claim 1, characterized in that: The membrane material characterization and membrane property analysis system at the membrane property detection end includes a contact angle meter, a permeation flux test system, an atomic force microscope, a zeta potential analyzer, and a thin film thermal conductivity test system, which can obtain the membrane contact angle, molecular weight cutoff, total charge, zeta potential, and thermal conductivity of the membrane material to be tested.

3. The membrane treatment performance prediction system based on machine learning according to claim 1, characterized in that: The pollutant characteristic analysis system at the pollutant property detection and calculation end includes a pollutant molecular weight collection system, a transmission electron microscope, an X-ray photoelectron spectrometer, a UV-Vis spectrophotometer, and a high performance liquid chromatograph, which can accurately measure the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of the pollutant; the pollutant descriptor calculation system is alvaDesc software, which can obtain 5290 descriptors that quantify molecular structure feature information.

4. The membrane treatment performance prediction system based on machine learning according to claim 1, characterized in that: The operating condition monitoring system of the operating condition collection end includes a multi-parameter online water quality detector, a rotor flowmeter, an osmometer, an electromagnetic flowmeter, an ion chromatograph, a UV-visible spectrophotometer, and a surface force meter, which can monitor the pH, temperature, flow, osmotic pressure, operating time, cross-flow velocity, ion concentration, natural organic matter concentration, and the interaction force between membrane and pollutants during the membrane treatment process.

5. The membrane treatment performance prediction system based on machine learning according to claim 1, characterized in that: The specific calculation method of the cloud computing learning end is: use the KNN algorithm to calculate the fourth nearest neighbor distance corresponding to each data point, draw the probability density distribution curve of the distance value, select the distance value corresponding to the inflection point of the curve as the domain radius in the DBSCAN algorithm; calculate the number of data points of each data point in the neighborhood range, and then calculate the closest distance of each point to a higher density point. If it is the highest density point, take the distance to the farthest point in the data set, and then use the number of data points in the neighborhood range as the horizontal coordinate and the closest distance to the higher density point as the vertical coordinate to draw a plot, and determine the range of the minimum number of points that a given neighborhood needs to include when noise points may exist; determine the minimum number of points that need to be included in different given neighborhoods and the selected domain radius to develop a DBSCAN model, and calculate the silhouette coefficient of the model, select the minimum number of points that need to be included in the neighborhood corresponding to the model with the silhouette coefficient closest to 1 as the optimal hyperparameter, and develop the optimal DBSCAN model to identify outliers in each variable data and replace them with missing values; Re-identify the missing values ​​of all variables and use the MissForest algorithm to fill the missing values; Then, the independent variables were screened and the model was developed: the SHAP value was used to measure the importance of the independent variables, and the independent variables were screened through cross-validation. The statistical method of machine learning was used to build a model to calculate and predict the membrane removal rate and water flux, and a model was developed to design and predict the balance point between the removal rate and water flux.

6. A prediction method of a membrane treatment performance prediction system based on machine learning as claimed in claim 1, characterized in that: The following steps are involved: Step 1) First, use a contact angle meter, a permeation flux test system, an atomic force microscope, a zeta potential analyzer, and a thin film thermal conductivity test system to analyze and test the membrane material to be tested, and obtain membrane property data including membrane contact angle, molecular weight cutoff, total charge, zeta potential, and thermal conductivity; Step 2) The wastewater to be treated is then sent to the pollutant property detection and calculation end, and the molecular weight, compound size, compound charge, water solubility, and octanol-water partition coefficient of the pollutants are measured using a molecular weight acquisition system, a transmission electron microscope, an X-ray photoelectron spectrometer, a UV-Vis spectrophotometer, and a high-performance liquid chromatograph; and 5290 molecular structure descriptors are calculated using alvaDesc software, integrated, and obtained pollutant characteristic data; Step 3) Afterwards, the pH, temperature, flow rate, osmotic pressure, operation time, cross flow velocity, ion concentration, natural organic matter concentration, and membrane-pollutant interaction force during the membrane treatment process are detected by a multi-parameter online water quality detector, a rotor flowmeter, an osmometer, an electromagnetic flowmeter, an ion chromatograph, a UV-visible spectrophotometer, and a surface force meter in the operation condition collection terminal to obtain membrane treatment operation data, and the membrane treatment operation data is recorded in real time; Step 4) The data obtained from the membrane characteristic detection end, the pollutant property detection and calculation end, and the operation condition collection end are then transmitted to the cloud computing learning end. The statistical method of machine learning is adopted, the computing resources of the cloud service computer cluster are utilized, and according to the statistical model and parameters, the optimization algorithm is adopted to calculate and predict the membrane removal rate and water flux of the constructed model on the cloud service computer cluster, and the model is developed to design and predict the balance point between the removal rate and the water flux; then the predicted data calculated by the cloud computing learning end is displayed on the data output end, and at the same time, the data output end can monitor the computing status of the cloud computing learning end and allocate computing cluster node resources according to the computing amount; Then the data output end feeds back the predicted membrane removal rate, water flux and the balance point between the removal rate and the water flux to the feedback correction end. The feedback correction end is connected to the big data system on the one hand to obtain the removal rate, water flux and the balance point data between the removal rate and the water flux of each pollutant under different conditions. At the same time, the feedback correction section is also connected to the experimental test bench to obtain the removal rate, water flux and the balance point between the removal rate and the water flux obtained by the actual test; The feedback correction end compares the predicted results with big data and actual experimental results, and feeds back the result deviation to the cloud computing learning end. The cloud computing learning end continuously corrects the predicted model and algorithm to obtain the optimal membrane treatment removal rate, water flux and the balance point prediction model between the removal rate and water flux, which is used for the prediction of membrane treatment removal rate, water flux and the balance point between the removal rate and water flux.

Images