Method for evaluating quality of photocatalytic membrane filtered water by using self-organizing mapping neural network

Abstract

The invention discloses a method for evaluating the quality of photocatalytic membrane filtered water by using a self-organizing mapping neural network, and belongs to the technical field of water treatment. According to the method, various water quality parameters such as UV254, dissolved organic carbon (DOC) and a three-dimensional fluorescence spectrum are integrated, and data are subjected to dimensionality reduction and clustering by using SOM, so that the water quality of the water filtered by various photocatalytic membrane filtration processes is evaluated more comprehensively and accurately, and an effective means is provided for comparison and optimization of the water quality of the water filtered by the photocatalytic membrane under different process conditions.

Description

Technical Field

[0001] The invention relates to the technical field of water treatment, and in particular to a method for evaluating the quality of water filtered out of a photocatalytic membrane by using a self-organizing mapping neural network. Background Art

[0002] Ultrafiltration is a membrane separation technology widely used in water treatment, which can effectively remove suspended particles, colloids and macromolecular organic matter in water. Ultrafiltration still has certain limitations in treating organic pollutants: on the one hand, ultrafiltration membranes mainly rely on physical interception to remove macromolecular organic matter, and have low removal efficiency for soluble small molecular organic matter. On the other hand, during long-term operation, a pollution layer composed of inorganic matter, organic matter and microorganisms is easily formed on the membrane surface, resulting in a decrease in membrane flux and reduced treatment efficiency. Photocatalytic membrane filtration technology loads photocatalysts (such as Bi2O3 / rGO / TiO2, Bi2O3-TiO2 / PAC, Bi2O3-TiO2 / mesoporous carbon) on the membrane surface to produce active oxygen species (such as hydroxyl radicals) under light conditions, thereby efficiently degrading organic pollutants in water. It has obvious membrane anti-pollution and self-cleaning capabilities, and can extend the service life of the membrane. However, how to accurately and comprehensively evaluate the effluent quality of photocatalytic membranes is still worthy of attention. Common membrane filtration effluent quality evaluation methods mainly focus on several key water quality indicators, such as UV 254 , dissolved organic carbon (DOC) and three-dimensional fluorescence spectral data, etc. The current water quality evaluation methods have certain limitations in reflecting the overall membrane filtration water quality status and process efficiency, and fail to effectively integrate these water quality indicators to form a comprehensive water quality evaluation system, especially for photocatalytic membrane filtration water after adsorption-oxidation-filtration coupling treatment. The existing methods lack effective multi-index evaluation means, and it is difficult to accurately evaluate and compare the pollution removal effects under different treatment process conditions. In recent years, self-organizing map neural network (SOM) as an unsupervised learning method has shown excellent performance in the field of pattern recognition and data classification. SOM can reduce the dimensionality and cluster high-dimensional data, realize the visualization and analysis of complex data, and is very suitable for processing multivariate and nonlinear water quality parameter data. By integrating UV 254 , DOC, three-dimensional fluorescence spectrum and other water quality parameters, SOM can provide more comprehensive and accurate water quality evaluation results, and realize data visualization, in-depth understanding and analysis of water quality changes under different process conditions. The present invention aims to provide a method for evaluating the water quality of photocatalytic membrane filtration using a self-organizing map neural network, to achieve accurate and comprehensive evaluation of photocatalytic membrane filtration water quality, and to select suitable process conditions. Summary of the invention

[0003] In view of the imperfect water quality evaluation method of photocatalytic membrane filtration water, the present invention proposes a method of evaluating the photocatalytic membrane treatment effect by using a self-organizing map neural network. 254 , dissolved organic carbon (DOC), three-dimensional fluorescence spectrum and other water quality parameters, and used SOM to reduce the dimension and cluster the data, so as to more comprehensively and accurately evaluate the water quality of effluents from various photocatalytic membrane filtration processes, providing an effective means for comparing and optimizing the water quality of photocatalytic membrane effluents under different process conditions.

[0004] The present invention adopts the following technical solutions:

[0005] A method for evaluating the water quality of photocatalytic membrane effluent using a self-organizing map neural network comprises the following steps:

[0006] (1) Prepare a multifunctional photocatalytic ceramic membrane with adsorption-oxidation-filtration coupling;

[0007] (2) fixing the photocatalytic ceramic membrane on a membrane reactor containing a water sample to be treated, performing overall photocatalytic filtration, and collecting effluent from the photocatalytic ceramic membrane filtration process under different treatment process conditions, and measuring the effluent quality under different treatment process conditions, wherein the water quality includes different indicators and data corresponding to the indicators, which are used for later evaluation of the treatment effect;

[0008] (3) storing the water quality index parameters and data collected in step (2) in an Excel table, and constructing a data set suitable for a self-organizing map neural network based on the data;

[0009] (4) Using MATLAB software, first read the Excel table provided in step (3); then use Fluor_Somap and SOM toolbox toolkits to import the Excel table data as input parameters into the unsupervised learning model, set the self-organizing map neural network parameters, and generate the final water quality evaluation results; preferably, the water quality evaluation results are in the form of a visualized BMU diagram, wherein the BMU diagram is a result of the selection of different water quality samples in the form of clusters filled into different polygonal structures of an array composed of polygonal structures corresponding to neurons, and a visualization result is formed after filling; at the same time, the above-mentioned BMU diagram is further classified into 2-10 levels of effects such as excellent, good, medium, medium, and poor, and the trend difference of each water quality index plane diagram is reflected.

[0010] The preparation of the photocatalytic ceramic membrane with adsorption-oxidation-filtration multifunctional coupling in step (1) further comprises the following steps:

[0011] 1) adding Bi2O3 / rGO / TiO2 or / and Bi2O3-TiO2 / PAC composites having both adsorption and oxidation functions into ultrapure water, and performing ultrasonic treatment for 20 min and magnetic stirring for 10 min to achieve uniform dispersion of the catalyst in the aqueous solution to form a suspension;

[0012] 2) depositing the composite material in the suspension onto the surface of an unmodified original ceramic membrane and drying it, for example, at 50° C.;

[0013] 3) placing the ceramic membrane coated with the composite material modified layer obtained in step 2) in an oven at 160° C. and heating it for 1 hour to enhance the bonding force between the composite material and the membrane surface;

[0014] Every 17.35cm 2 The area of ​​the original ceramic membrane surface corresponds to the mass of Bi2O3 / rGO / TiO2 and Bi2O3-TiO2 / PAC composites of 0-100 mg.

[0015] Furthermore, the overall photocatalytic filtration method of step (2) is set as follows:

[0016] 1) The light sources used for the photocatalytic ceramic membrane include visible light, full-band light and ultraviolet light, and the light power density irradiated to the membrane surface is set to 200-500mW / cm 2 ;

[0017] 2) Fixing the photocatalytic ceramic membrane on a membrane reactor containing the water sample to be treated, using a cross-flow filtration mode, setting the filtration pressure to 0.05-0.2 MPa, and the cross-flow rate to 50-200 mL / min;

[0018] Furthermore, the different treatment processes are divided into dark treatment, dark plus light treatment, and light treatment;

[0019] 1) Dark treatment refers to fixing the photocatalytic ceramic membrane on a membrane reactor containing the liquid to be treated and filtering it under no light conditions;

[0020] 2) Dark plus light treatment refers to fixing the photocatalytic ceramic membrane on the membrane reactor containing the liquid to be treated, filtering for a period of time under no light conditions, and then turning on the light source for filtering;

[0021] 3) Light treatment refers to fixing the photocatalytic ceramic membrane on the membrane reactor containing the liquid to be treated, and filtering is always carried out under the condition of turning on the light source;

[0022] 4) Dark treatment, dark plus light treatment, and light treatment also include different treatment times;

[0023] The water quality indicators measured include:

[0024] 1) Use UV spectrophotometer to measure the UV of the effluent from the photocatalytic ceramic membrane 254 Value to indicate the concentration of organic matter in the water sample;

[0025] 2) The dissolved organic carbon (DOC) value of the effluent from the photocatalytic ceramic membrane was measured by a total organic carbon analyzer to evaluate the total amount of dissolved organic matter in the water sample;

[0026] 3) Under specific conditions (excitation wavelength range 200-450nm, step length 5nm; emission wavelength range 250-550nm, step length 2nm; scanning rate 1200nm / min), the three-dimensional fluorescence spectrum of the effluent from the photocatalytic ceramic membrane was measured;

[0027] 4) The 3D fluorescence spectrum data obtained in step 3) were processed using the DOMFlour toolkit in MATLAB. All fluorescence peaks were first converted into Raman units (RU). A binary search algorithm and a random initialization strategy were applied to verify and determine the number of effective fluorescent components (2-7) and the maximum fluorescence intensity (F) of each component. max );

[0028] Furthermore, the data set includes the following three Excel tables:

[0029] 1) The first table is defined as an input data matrix, which is the specific value of each water quality index corresponding to each water quality sample, with m rows and n columns, where m represents the number of samples, a row represents the tested water quality samples (including raw water samples and water quality samples after different treatment processes), a row represents a sample, n represents the number of types of water quality index parameters, and a column represents different water quality index parameters, and a column represents a water quality index parameter, which is used to input the neural network for data processing;

[0030] 2) The second table is a single column or a single row, listing the numbers of m samples, i.e., the numbers of different operation processes, where each number corresponds to a different operation process of a water quality sample in a row in the first table;

[0031] 3) The third table is a single column or a single row, listing n water quality index parameter names, and each water quality index parameter name corresponds to a column of water quality indicators in the first table in turn.

[0032] Furthermore, the principle of visualization of the results in step (4) using Fluor_Somap and SOM toolbox includes:

[0033] 1) The input data is compared with the neuron weights on the map through a competition mechanism to find the best matching neuron, as shown in formula (1);

[0034] ||xm c||=min{||xm i ||} (1)

[0035] 2) According to the input data, the weights of the winning neuron and its neighboring neurons are adjusted so that similar data tend to cluster on the map, as shown in formula (2); by iterating this process, the orderly distribution of data on the map is achieved.

[0036] m i (t+1)=m i (t)+α(t)·N c (t)·[x(t)-m i (t)] (2)

[0037] Where x is the input vector, m i is the map of all mapping vectors, m c is the closest mapping vector to the input vector, ||x|| is the Euclidean distance, m i (t+1) is the new weight of the neuron, m i (t) is the previous weight of the neuron, α(t) is the learning rate, and N C is the neighborhood size of a neuron.

[0038] Furthermore, the model parameter settings include:

[0039] 1) Determine the number of neurons l according to the number of samples m, as shown in formula (3);

[0040]

[0041] 2) The number of linear initialization maps can be selected as needed, generally 1-10; the number of random initialization maps can be selected as needed, generally 10-50; the number of iterations for rough adjustment of the network can be selected as needed, generally 100-500; the number of iterations for fine adjustment of the network can be selected as needed, generally 5000-10000, the learning rate can be selected as needed, generally 0.001-0.01, and the neighborhood radius can be selected as needed, generally 2-4;

[0042] 3) After the calculation is completed, the map with the lowest mean quantization error (mqe) and topological grid error (tge) is selected for export.

[0043] Furthermore, the evaluation results of the effluent quality of the photocatalytic ceramic membrane include:

[0044] (1) A self-organizing map neural network (SOM) was used to evaluate the degradation performance of organic pollutants in effluent water. The variables input into the neural network were water quality index parameters, including UV 254, dissolved organic carbon (DOC), and 2-7 fluorescent components obtained by parallel factor analysis and the F of each component max Value (for example, if there are three fluorescent components, then the F values ​​of the three fluorescent components and each component are max The values ​​correspond to F of C1, C2, and C3. max value); SOM uses Euclidean distance as a metric to judge the similarity between samples; UV 254 , DOC and the F of each component corresponding to 2-7 fluorescent components max The values ​​are usually high, while the values ​​of these indicators of the effluent from the process with the best treatment effect are relatively low. Therefore, the Euclidean distance between the raw water and the effluent with the best treatment effect is larger; on the contrary, the Euclidean distance between the effluent with poor treatment effect and the raw water is smaller; this difference based on Euclidean distance provides us with an effective means to evaluate the effects of different treatment processes.

[0045] (2) Based on the above principle, a best matching unit (BMU) graph is generated, for example Fig.10 , which is used to visually display the effluent effects of different treatment processes and treatment times; in the BMU map, the raw water samples are located at the bottom (or top) of the map, and the effluent samples with poor treatment effects are close to the raw water samples. The values ​​of various indicators of these samples are high, indicating that the removal effect of organic pollutants is not good; the effluent samples with good treatment effects are distributed at the top (or bottom) of the map, and the UV values ​​of these samples are 254 , DOC and the F of each component corresponding to 2-7 fluorescent components max The lower the value, the better the degradation of organic pollutants;

[0046] 3) The effluent samples in the BMU map are divided into 2 to 10 clusters according to similarity (can be selected as needed). The generated cluster map can further classify the water quality in the BMU map into 2-10 levels such as excellent, good, medium, average, and poor; for example Fig.11 wait.

[0047] 4) The generated component (each water quality index parameter) plane diagram can intuitively show the trend differences of different indicators under different treatment processes, such as Figure 12-16 .

[0048] The beneficial effects of the present invention are:

[0049] (1) The present invention successfully constructed a multifunctional coupled photocatalytic ceramic membrane with adsorption-oxidation-filtration properties by coating photocatalytic composite materials such as Bi2O3 / rGO / TiO2 and Bi2O3-TiO2 / PAC on the surface of a ceramic ultrafiltration membrane. Compared with unmodified ceramic ultrafiltration membranes, the photocatalytic membrane exhibits better performance in removing organic matter and has significant anti-pollution ability.

[0050] (2) The present invention proposes a multi-index effluent water quality evaluation method, which includes UV 254 , DOC and three-dimensional fluorescence spectrum data are used as input variables of the self-organizing map neural network, and the water quality effect is evaluated by the difference in Euclidean distance. This method can reduce the dimension of complex high-dimensional data to two-dimensional space and realize the visualization processing of multivariate data.

[0051] (3) The present invention generates the best matching unit and component plane diagrams through a self-organizing map neural network. These diagrams can intuitively show the impact of different treatment process conditions on the water quality of the photocatalytic membrane effluent, as well as the differences in the removal effects of different water quality indicators, and select the appropriate process conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The UV of the secondary biochemical effluent was filtered by CUM and MCUM-50 in the specific embodiment under dark, dark-UV and UV conditions. 254 Change graph;

[0053] Figure 2 It is a graph showing the DOC changes of CUM and MCUM-50 in a specific embodiment after filtering the secondary biochemical effluent under dark, dark-UV and UV conditions;

[0054] Figure 3 The fluorescence morphology of component C1 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific embodiment under dark, dark-UV and UV conditions;

[0055] Figure 4 The excitation and emission spectral characteristics of component C1 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific implementation manner under dark, dark-UV and UV conditions;

[0056] Figure 5 The fluorescence morphology of component C2 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific embodiment under dark, dark-UV and UV conditions;

[0057] Figure 6 The excitation and emission spectral characteristics of component C2 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific implementation manner under dark, dark-UV and UV conditions;

[0058] Figure 7 The fluorescence morphology of component C3 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific embodiment under dark, dark-UV and UV conditions;

[0059] Figure 8 The excitation and emission spectral characteristics of component C3 analyzed by the parallel factor method after filtering the secondary biochemical effluent of CUM and MCUM-50 in the specific implementation manner under dark, dark-ultraviolet and ultraviolet conditions;

[0060] Fig. 9 The Fmax changes of three fluorescent components in the secondary biochemical effluent at different filtration times were obtained by filtering the secondary biochemical effluent with CUM and MCUM-50 in the specific implementation manner under dark, dark UV and UV conditions.

[0061] Fig.10 The BMU diagram generated by SOM in a specific implementation manner is shown.

[0062] Fig.11 A cluster diagram of a BMU map generated by SOM in a specific implementation is shown.

[0063] Fig.12 A component plan view of component C1 generated by SOM in a specific embodiment is shown.

[0064] Fig.13 A component plan view of component C2 generated by SOM in a specific embodiment is shown.

[0065] Fig.14 A component plan view of component C3 generated by SOM in a specific embodiment is shown.

[0066] Fig.15 Shows the UV generated by SOM in a specific implementation 254 Component plan.

[0067] Fig.16 A component plan view of the DOC generated by the SOM in a specific embodiment is shown.

[0068] Explanation of the attached table

[0069] Table 1 shows the first table of the SOM input dataset in the specific implementation

[0070] Table 2 shows the second table of the SOM input dataset in the specific implementation

[0071] Table 3 shows the third table of the SOM input dataset in the specific implementation DETAILED DESCRIPTION

[0072] (1) Prepare a multifunctional photocatalytic ceramic membrane with adsorption-oxidation-filtration coupling;

[0073] (2) fixing the photocatalytic ceramic membrane on a membrane reactor containing a water sample to be treated, collecting effluent from the photocatalytic ceramic membrane filtration process under different working conditions and treatment processes, and measuring a series of water quality indicators to evaluate the treatment effect;

[0074] (3) storing the water quality parameter data collected in step (2) in an Excel table, and constructing a data set suitable for a self-organizing map neural network based on the data;

[0075] (4) Using MATLAB software, first read the table data provided in step (3). Then, through Fluor_Somap and SOM toolbox, import these data as input parameters into the unsupervised learning model, set the model parameters, and generate the final water quality evaluation results.

[0076] The present invention is described in detail by the following examples, which are only part of the examples of the present invention, rather than all of the embodiments.

[0077] The photocatalytic ceramic membrane has a diameter of 47 mm and a surface area of ​​17.35 cm 2 The mass of the Bi2O3 / rGO / TiO2 mixture deposited on the flat commercial ceramic ultrafiltration membrane is 50 mg.

[0078] The working condition described is that the photocatalytic ceramic membrane uses ultraviolet light as the light source, adopts the five-point sampling method, and sets the light power density irradiated to the membrane surface to 400mW / cm 2 The membrane filtration adopted the cross-flow filtration mode, the filtration pressure was set to 0.1Mpa, the cross-flow rate was set to 100mL / min, the total filtration time was 120min, and the sampling interval was 15min.

[0079] The photocatalytic ceramic membrane filtration process includes dark treatment, dark plus light treatment, and light treatment.

[0080] The water quality indicators include UV 254 , DOC, and F of three fluorescent components max (Corresponding to F of C1, C2, C3 max ).

[0081] The data set includes three Excel tables. The first table is defined as an input data matrix, with 49 rows and 5 columns, that is, it has 49 sample numbers and 5 water quality parameter types. This table is used to input the neural network for data processing; the second table contains the numbers of the 49 samples, each number corresponds to a sample in the first table; the third table lists the parameter names of the 5 water quality parameters, namely Component 1, Component 2, Component 3, UV 254, DOC. Specific implementation case tables are shown in Appendix 1, Appendix 2 and Appendix 3.

[0082] The model parameters are set with 36 neurons (corresponding to Fig.10 36 hexagonal units in the network); set the number of linear initialization maps to 1; the number of random initialization maps to 10; the number of iterations for rough adjustment of the network to 100; the number of iterations for fine adjustment of the network to 5000, the learning rate to 0.001, the winning neighborhood radius to 3, and after the calculation is completed, select the map with the lowest average quantization error (mqe) and topological mesh error (tge) to export the results (corresponding to the attached Figure 10-16 , where each row in each picture is 3 hexagons, each column is 12 hexagons, a total of 36, corresponding to 36 neurons).

[0083] The evaluation results include:

[0084] (1) Generate a best matching unit (BMU) map to visually display the effluent effects of different treatment processes and treatment times. In the BMU map, the raw water samples are located at the bottom of the map, and the effluent samples with poor treatment effects are close to the raw water samples. The values ​​of various indicators of these samples are high, indicating that the removal effect of organic pollutants is poor. The effluent samples with good treatment effects are distributed at the top of the map. The UV values ​​of these samples are 254 , DOC and F of fluorescent components max The lower the value, the better the degradation effect of organic pollutants.

[0085] (2) The generated cluster map divides the effluent samples in the BMU map into 6 clusters based on similarity.

[0086] (3) The generated component plane diagram can intuitively show the trend differences of different treatment processes and treatment times in the degradation of different indicators.

[0087] The present invention is described in detail below through specific embodiments and drawings. The drawings constitute a part of the present invention and are used to illustrate the principles of the present invention rather than to limit the scope of the present invention.

[0088] Embodiment 1:

[0089] (1) A photocatalytic ceramic ultrafiltration membrane (MCUM-50) with a coating mass of 50 mg was prepared, and an unmodified ceramic ultrafiltration membrane (CUM) was used as a comparison.

[0090] (2) Under the set working conditions, fix the MCUM-50 or CUM on the UV 254 The value is 0.1311cm-1, and the DOC concentration is 4.254mg / L. The F of components C1, C2, and C3 maxThe membrane reactor with values ​​of 0.87, 1.04, and 0.47 RU, respectively, was used for filtration experiments under dark treatment, dark plus light treatment, and light treatment conditions. The filtration time was 120 min, and the permeate was collected every 15 min. Determination of basic water quality parameters (UV 254 , DOC, three-dimensional fluorescence). The three-dimensional fluorescence spectrum uses the DOMFlour toolbox in MATLAB to perform parallel factor analysis on the fluorescence spectrum data, and the number of effective fluorescent components and the maximum fluorescence intensity F of each fluorescent component are obtained by dichotomy and random assignment test. max After analysis, three fluorescent components were obtained. The specific results of each component are shown in Table 1 and Figure 3-8 .

[0091] Table 1. Characteristics of three fluorescent components in organic pollutants determined by parallel factor analysis

[0092]

[0093]

[0094] (3) The water quality parameters (UV 254 , DOC, and F of the three fluorescent components max The data are stored in Excel tables. Figure 1 and Figure 2 UV 254 and DOC trends, where lower values ​​indicate better water quality; Fig. 9 The F values ​​of the three fluorescent components are shown in a stacked bar graph. max The value changes. In the figure, water samples were collected at 15, 30, and 120 minutes of filtration. The sample process numbers are shown in Table 2. In addition, the shorter columns in the figure represent lower content of fluorescent components, which also indicates good water quality. Based on these data, a data set suitable for self-organizing map neural network analysis was produced. The data set tables are shown in Appendix 1, Appendix 2, and Appendix 3;

[0095] (4) Read the table in step 3, use the Fluor_Somap and SOM toolbox toolkits in MATLAB, input the set model parameters into the unsupervised learning model for calculation, and generate water quality evaluation results. Among them, CUM and MCUM-50 filtered the secondary biochemical effluent under dark, dark ultraviolet and ultraviolet conditions. The naming and numbering of the effluents at different filtration times are shown in Table 2.

[0096] Table 2

[0097]

[0098]

[0099] Table 3 Excel tables

[0100]

[0101]

[0102]

[0103] Fig.10 The Best Matching Unit (BMU) diagram shown in the figure reveals the effects of different treatment processes and time on the effluent effect. By clustering the samples on the BMU diagram, we divided the treatment effect areas, such as Fig.11 As shown in Figure 1, a total of six clusters were formed. Among these clusters, cluster 3 represents the worst treatment effect area, cluster 1 is the poor treatment area, clusters 4, 5 and 6 are classified as medium treatment areas, and the samples in cluster 2 indicate good treatment effects. Fig.10 In the figure, the raw water sample (SE) is located at the bottom of the map, while the effluent samples with poor treatment effects (such as CUM19, CUM20, CUM22, CUM23, and CUM24) are next to it and belong to cluster 3. The index values ​​of these samples are high, reflecting the unsatisfactory removal of organic pollutants, indicating that single ultrafiltration and secondary biochemical treatment have limited effect on improving water quality. On the contrary, the effluent samples with significant treatment effects (such as MCUM4, MCUM5, MCUM6, MCUM7, MCUM8, etc.) are concentrated in the cluster 2 area at the top of the map. The UV values ​​of these samples are 254 Absorbance, dissolved organic carbon (DOC) content and maximum fluorescence intensity (F max ) were relatively low, proving that photocatalytic membrane technology can effectively improve water quality, and with the increase of filtration time, the degradation efficiency of organic matter is further improved.

[0104] Without UV light, most samples (CUM9-10, CUM17-24, MCUM20-24) clustered in the lower half of the BMU graph, reflecting that the secondary biochemical effluent underwent similar levels of treatment under these conditions. When UV light was turned on, most samples migrated to the upper half of the BMU graph, indicating that the direct photolysis of UV light significantly enhanced the removal of organic pollutants in the secondary biochemical effluent. In particular, MCUM-50 showed superior pollutant removal compared to the CUM series, which was attributed to the reactive oxygen species generated by Bi2O3 / rGO / TiO2 on the surface of MCUM-50 under UV irradiation, which had a highly efficient degradation effect on organic pollutants.

[0105] Figures 12 to 16 For simulated C1 to C3, UV 254, DOC component plane diagrams, which provide an intuitive view of the changing trends of various water quality indicators under different treatment processes and treatment times. The component plane diagrams of C1 to C3 gradually decrease from the lower left (representing raw water) to the top, indicating that the photocatalytic ceramic membrane technology performs well in removing fluorescent substances. It is worth noting that the changing trends of C1 and C2 components are similar, suggesting that these two types of substances may have similar degradation characteristics; similarly, C3 and UV 254 The degradation trends of DOC and DOC are similar, while the degradation paths of DOC and other water quality indicators are different. This finding emphasizes that the photocatalytic ceramic membranes have different removal characteristics for different types of water quality indicators.

[0106] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating the water quality of photocatalytic membrane effluent using a self-organizing map neural network, characterized in that: The following steps are involved: (1) Prepare a multifunctional photocatalytic ceramic membrane with adsorption-oxidation-filtration coupling; (2) fixing the photocatalytic ceramic membrane on a membrane reactor containing a water sample to be treated, performing overall photocatalytic filtration, and collecting effluent from the photocatalytic ceramic membrane filtration process under different treatment process conditions, and measuring the effluent quality under different treatment process conditions, wherein the water quality includes different indicators and data corresponding to the indicators, which are used for later evaluation of the treatment effect; (3) storing the water quality index parameters and data collected in step (2) in an Excel table, and constructing a data set suitable for a self-organizing map neural network based on the data; (4) Using MATLAB software, first read the Excel table provided in step (3); then, through the Fluor_Somap and SOM toolbox toolkits, import the Excel table data as input parameters into the unsupervised learning model, set the self-organizing map neural network parameters, and generate the final water quality evaluation results.

2. The method according to claim 1, characterized in that The water quality evaluation result is preferably in the form of a visualized BMU diagram, wherein the BMU diagram is a result of filling different water quality samples into different polygonal structures of an array composed of polygonal structures corresponding to neurons in the form of clusters, and forming a visualized result after filling; The water quality evaluation results also include the effect of further classifying the BMU map into 2-10 levels, and reflecting the trend differences of each water quality indicator plane map.

3. The method according to claim 1, characterized in that The preparation of the photocatalytic ceramic membrane with multifunctional coupling of adsorption-oxidation-filtration in step (1) comprises the following steps: 1) adding Bi2O3 / rGO / TiO2 or / and Bi2O3-TiO2 / PAC composites having both adsorption and oxidation functions into ultrapure water, and performing ultrasonic treatment for 20 min and magnetic stirring for 10 min to achieve uniform dispersion of the catalyst in the aqueous solution to form a suspension; 2) depositing the composite material in the suspension onto the surface of an unmodified original ceramic membrane and drying it, for example, at 50° C.; 3) placing the ceramic membrane coated with the composite material modified layer obtained in step 2) in an oven at 160° C. and heating it for 1 hour to enhance the bonding force between the composite material and the membrane surface; Every 17.35cm 2 The area of ​​the original ceramic membrane surface corresponds to the mass of Bi2O3 / rGO / TiO2 and Bi2O3-TiO2 / PAC composites of 0-100 mg.

4. The method according to claim 1, characterized in that Step (2) The overall photocatalytic filtration method is set up as follows: 1) The light sources used for the photocatalytic ceramic membrane include visible light, full-band light and ultraviolet light, and the light power density irradiated to the membrane surface is set to 200-500mW / cm 2 ; 2) Fixing the photocatalytic ceramic membrane on a membrane reactor containing the water sample to be treated, using a cross-flow filtration mode, setting the filtration pressure to 0.05-0.2 MPa, and the cross-flow rate to 50-200 mL / min; The different treatment processes are divided into dark treatment, dark plus light treatment, and light treatment; 1) Dark treatment refers to fixing the photocatalytic ceramic membrane on a membrane reactor containing the liquid to be treated and filtering it under no light conditions; 2) Dark plus light treatment refers to fixing the photocatalytic ceramic membrane on the membrane reactor containing the liquid to be treated, filtering for a period of time under no light conditions, and then turning on the light source for filtering; 3) Light treatment refers to fixing the photocatalytic ceramic membrane on the membrane reactor containing the liquid to be treated, and filtering is always carried out under the condition of turning on the light source; 4) Dark treatment, dark plus light treatment, and light treatment also include different treatment times; The water quality indicators measured include: 1) Use UV spectrophotometer to measure the UV of the effluent from the photocatalytic ceramic membrane 254 Value to indicate the concentration of organic matter in the water sample; 2) The dissolved organic carbon (DOC) value of the effluent from the photocatalytic ceramic membrane was measured by a total organic carbon analyzer to evaluate the total amount of dissolved organic matter in the water sample; 3) Under specific conditions (excitation wavelength range 200-450nm, step length 5nm; emission wavelength range 250-550nm, step length 2nm; scanning rate 1200nm / min), the three-dimensional fluorescence spectrum of the effluent from the photocatalytic ceramic membrane was measured; 4) The 3D fluorescence spectral data obtained in step 3) were processed using the DOMFlour toolkit in MATLAB. All fluorescence peaks were first converted into Raman units (RU). A binary search algorithm and a random initialization strategy were applied to verify and determine the number of effective fluorescent components (2-7) and the maximum fluorescence intensity (F) of each component. max ).

5. The method according to claim 1, characterized in that Step (3) The data set includes the following three Excel tables: 1) The first table is defined as the input data matrix, which is the specific value of each water quality index corresponding to each water quality sample, with m rows and n columns, where m represents the number of samples, the row represents the water quality sample, a row represents one sample, n represents the number of types of water quality index parameters, the column represents different water quality index parameters, and one column represents one water quality index parameter, which is used to input the neural network for data processing; 2) The second table is a single column or a single row, listing the numbers of m samples, i.e., the numbers of different operation processes, where each number corresponds to a different operation process of a water quality sample in a row in the first table; 3) The third table is a single column or a single row, listing n water quality index parameter names, and each water quality index parameter name corresponds to a column of water quality indicators in the first table in turn.

6. The method according to claim 1, characterized in that Step (4) Fluor_Somap and SOM toolbox implement the result visualization principle including: 1) The input data is compared with the neuron weights on the map through a competition mechanism to find the best matching neuron, as shown in formula (1); ||x-m c ||=min{||x-m i ||} (1) 2) According to the input data, the weights of the winning neuron and its neighboring neurons are adjusted so that similar data tend to cluster on the map, as shown in formula (2); by iterating this process, the orderly distribution of data on the map is achieved. m i (t+1)=m i (t)+α(t)·N c (t)·[x(t)-m i (t)] (2) Where x is the input vector, m i is the map of all mapping vectors, m c is the closest mapping vector to the input vector, ||x|| is the Euclidean distance, m i (t+1) is the new weight of the neuron, m i (t) is the previous weight of the neuron, α(t) is the learning rate, and N C is the neighborhood size of a neuron.

7. The method according to claim 1, characterized in that The model parameter settings include: 1) Determine the number of neurons l according to the number of samples m, as shown in formula (3); 2) The number of linear initialization maps can be selected as needed, generally 1-10; the number of random initialization maps can be selected as needed, generally 10-50; the number of iterations for rough adjustment of the network can be selected as needed, generally 100-500; the number of iterations for fine adjustment of the network can be selected as needed, generally 5000-10000, the learning rate can be selected as needed, generally 0.001-0.01, and the neighborhood radius can be selected as needed, generally 2-4; 3) After the calculation is completed, the map with the lowest mean quantization error (mqe) and topological grid error (tge) is selected for export.

8. The method according to claim 1, characterized in that The evaluation results of the effluent quality of the photocatalytic ceramic membrane specifically include: (1) A self-organizing map neural network (SOM) was used to evaluate the degradation performance of organic pollutants in effluent water. The variables input into the neural network were water quality index parameters, including UV 254 , dissolved organic carbon (DOC), and 2-7 fluorescent components obtained by parallel factor analysis and the F of each component max value; SOM uses Euclidean distance as a metric to judge the similarity between samples; UV of raw water 254 , DOC and the F of each component corresponding to 2-7 fluorescent components max The values ​​are usually high, while the values ​​of these indicators of the effluent from the process with the best treatment effect are relatively low. Therefore, the Euclidean distance between the raw water and the effluent with the best treatment effect is larger; on the contrary, the Euclidean distance between the effluent with poor treatment effect and the raw water is smaller; this difference based on Euclidean distance provides us with an effective means to evaluate the effects of different treatment processes. (2) Based on the above principle, a best matching unit (BMU) map is generated to intuitively display the effluent effects of different treatment processes and treatment times. In the BMU map, the raw water samples are located at the bottom (or top) of the map, and the effluent samples with poor treatment effects are close to the raw water samples. The values ​​of various indicators of these samples are high, indicating that the removal effect of organic pollutants is poor. The effluent samples with good treatment effects are distributed at the top (or bottom) of the map. The UV values ​​of these samples are 254 , DOC and the F of each component corresponding to 2-7 fluorescent components max The lower the value, the better the degradation of organic pollutants; 3) The effluent samples in the BMU map are divided into 2 to 10 clusters according to similarity. The generated cluster map can further classify the water quality samples in the BMU map to form a 2-10 level effect; 4) The generated component plane diagram can intuitively show the trend differences of different indicators under different treatment processes.

Images