Online intelligent early warning and prevention system for iodine disinfection by-products in water plants

Abstract

The present invention discloses an online intelligent early warning and prevention and control system for iodine disinfection byproducts in a water supply plant, the system comprising an online halogen detection unit, an online fluorescence detection unit, a front ultraviolet disinfection unit, a rear chlorination disinfection unit, and an intelligent control unit, and is integrated and docked with an existing water plant control system. The present invention adopts a machine learning algorithm to construct an intelligent control unit, and by receiving the halogen level and the fluorescence level of dissolved organic matter of the raw water of an urban water supply plant, especially the offshore urban water supply plant during saltwater intrusion, through the operation process of the machine learning algorithm, a prediction effect of a determination coefficient of more than 0.90 is achieved, and the parameters of the ultraviolet first and then chloramine disinfection process for high iodine water are intelligently regulated, which can reduce the generation of iodine disinfection byproducts by 70.2%, and can be integrated and docked into the existing control system of the water supply plant as an expansion module for operation and control, and low-risk effluent with low iodine disinfection byproduct generation is obtained intelligently, efficiently and safely.

Description

Technical Field

[0001] The invention belongs to the technical field of water treatment and relates to a new method and system for online intelligent early warning and prevention of iodine disinfection by-products in a water supply plant. Background Art

[0002] Disinfection byproducts (DBPs) formed during water treatment have long attracted much attention due to their complex genotoxic and cytotoxic effects. The formation of DBPs is a byproduct of the reaction between disinfectants and dissolved organic matter, anthropogenic pollutants, iodides or bromides in water. In recent years, people have paid more attention to toxic iodinated DBPs. Studies have shown that iodinated DBPs exhibit higher levels of cytotoxicity, genotoxicity and developmental toxicity in mammalian cell assays compared with general chlorinated or brominated DBPs (Environmental Science & Technology 42 (22), 8330-8338).

[0003] In water, iodide ions (I - ), iodate (IO3 - ) and organic iodides are the main sources of iodine in water. They are likely to transform into each other during chlorination and chloramination disinfection, leading to the formation of iodinated disinfection by-products. Especially in the context of saltwater intrusion, the iodine content of raw water in coastal urban water plants will increase significantly, and the corresponding risk of iodinated disinfection by-products will also increase. Dissolved organic matter is the main precursor of iodinated disinfection by-products, a complex mixture of thousands of compounds. Ultraviolet irradiation has an excellent disinfection effect of inactivating waterborne pathogens, and its application in water plants is becoming more and more widespread. Due to the lack of long-lasting disinfection ability of ultraviolet rays, traditional disinfectants such as commonly used chloramines are still necessary. However, after ultraviolet irradiation, water containing both iodate and iodide ions forms significantly higher iodinated disinfection by-products during post-chloramination disinfection than water containing only iodate or only iodide ions (Water Res.193,116851).

[0004] Due to the complexity of the composition of soluble organic matter in water and the mutual conversion between iodide ions and iodates, it is difficult to accurately judge and control the production of iodine disinfection by-products in the process of UV-then-chloramine disinfection. Therefore, there is currently no technology that can predict and warn iodine disinfection in the process of UV-then-chloramine disinfection, especially in the context of saltwater intrusion into water quality, and there is no corresponding regulatory system application. Therefore, there is an urgent need for corresponding technical means to predict, warn, regulate and control iodine disinfection by-products. Summary of the invention

[0005] The present invention discloses an online intelligent early warning and prevention and control system for iodine disinfection byproducts in a water supply plant, the system comprising an online halogen detection unit, an online fluorescence detection unit, a front ultraviolet disinfection unit, a rear chlorination disinfection unit, and an intelligent control unit, and is integrated and docked with an existing water plant control system. The present invention adopts a machine learning algorithm to construct an intelligent control unit, and by receiving the halogen level and the fluorescence level of dissolved organic matter of the raw water of an urban water supply plant, especially the offshore urban water supply plant during saltwater intrusion, through the operation process of the machine learning algorithm, a prediction effect of a determination coefficient of more than 0.90 is achieved, and the parameters of the ultraviolet first and then chloramine disinfection process for high-iodine water are intelligently regulated, which can reduce the generation of iodine disinfection byproducts by 70.2%, and can be integrated and docked into the existing control system of the water supply plant as an expansion module for operation and control, and low-iodine disinfection byproducts and low-risk effluent are obtained intelligently, efficiently and safely.

[0006] To achieve the above objectives, the system comprises a halogen online detection unit (1), a fluorescence online detection unit (2), a pre-ultraviolet disinfection unit (3), a post-chlorination disinfection unit (4), and an intelligent control unit (5), and is integrated with an existing water plant control system (6); when the system treats iodine-containing water, first the halogen online detection unit (1) performs online detection of the concentrations of iodine ions, iodate ions and bromide ions in the water, and sends the concentration data to the intelligent control unit (5), while the fluorescence online detection unit (2) performs online detection of the three-dimensional fluorescence level (excitation emission matrix) of dissolved organic matter in the water, and sends the result to the intelligent control unit (5), while the pre-ultraviolet disinfection unit (3) and the post-chlorination disinfection unit (4) that have not yet been disinfected send the currently running disinfection parameters of ultraviolet light intensity and chlorination dosage to the intelligent control unit (5), and the intelligent control unit (5) performs online detection according to the machine learning algorithm model that has been written The iodine disinfection by-products to be generated under the current disinfection parameters are predicted and compared with the iodine disinfection by-product limit values ​​set by the water supply plant to determine whether the disinfection parameters need to be adjusted. Then, the front ultraviolet disinfection unit (3) and the rear chlorination disinfection unit (4) are controlled with the new disinfection parameters with low iodine disinfection by-product generation risk, so that the output of iodine disinfection by-products in the effluent is reduced, thereby reducing the corresponding risk. The machine learning algorithm model written into the intelligent control unit (5) is a machine learning algorithm that can be selected when the model is established, including extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm, random forest (RF) algorithm. At the same time, the model used should be the optimal parameter model trained by the water supply plant. The generation of the optimal parameter model is achieved by adjusting the hyperparameters on the training data set after segmenting the complete data set, and performing a test set model test to achieve a determination coefficient R on the test data set. 2The model with a value of ≥0.9 is finally applied, and the input variables of the complete training data set include the halogen level in water (iodine ion, iodate ion and bromide ion concentration), the fluorescence level of dissolved organic matter (decomposed excitation emission matrix signal), and the pre-ultraviolet-post-chlorination disinfection parameters, and the output variable is the level of iodine disinfection by-products under the corresponding input conditions; when applied, the intelligent control unit (5) receives the data from the halogen online detection unit (1), the fluorescence online detection unit (2), the pre-ultraviolet disinfection unit (3) and the post-chlorination disinfection unit (4) and performs predictive calculation and logical judgment. The process is to first predict the level of iodine disinfection by-products under the current disinfection parameters. If it is less than the iodine disinfection by-products set by the water plant, the level of iodine disinfection by-products is predicted. If the iodine disinfection by-product limit is greater than the iodine disinfection by-product limit set by the water plant, the series of disinfection parameters are cyclically input into the model for prediction, and the disinfection parameters of the front ultraviolet disinfection unit (3) and the rear chlorine disinfection unit (4) are reset to perform disinfection with the disinfection parameter whose prediction result is less than the iodine disinfection by-product limit set by the water plant. If all disinfection parameters after the cyclic input have no prediction results less than the iodine disinfection by-product limit set by the water plant, the disinfection parameters of the front ultraviolet disinfection unit (3) and the rear chlorine disinfection unit (4) are reset to perform disinfection with the disinfection parameter whose iodine disinfection by-product generation level is the minimum, and the data is fed back to the water plant control system (6) for early warning.

[0007] Specifically, the machine learning algorithm model written into the intelligent control unit (5) is an optimal parameter model. The generation and selection of the optimal parameter model includes the following steps or a partial combination of the following steps: ① Establishing a model with historical monitoring data of a water supply plant using the system, or open source water quality experimental data. The input variables of the complete data set used in the training process include three types of water halogen levels, dissolved organic matter fluorescence levels, and pre-ultraviolet-post-chlorination disinfection parameters. The halogen levels are specifically iodide ion, iodate ion and bromide ion concentrations. The dissolved organic matter fluorescence level is a decomposed excitation emission matrix signal. The decomposition method selected is selected from the parallel factor method, the principal component analysis method or the pre-machine learning method. The pre-ultraviolet-post-chlorination disinfection parameters include ultraviolet light intensity and chlorine. amine concentration; ② the output variable is the level of iodinated disinfection by-products under the corresponding input conditions, wherein the level of iodinated disinfection by-products uses the total iodinated disinfection by-product concentration according to the monitoring needs, or uses the overall index of total iodinated trihalomethanes and total iodinated haloacetic acid concentrations according to the monitoring needs, or uses a single or several specific iodinated disinfection by-product concentrations, and the total iodinated disinfection by-products include triiodomethane, monochlorodiiodomethane, dichloroiodomethane, monobromodiiodomethane, dibromoiodomethane, chlorobromoiodomethane, as well as monoiodoacetic acid, diiodoacetic acid or triiodoacetic acid; ③ the extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm or random forest (RF) algorithm are used for training, adjustment and testing at the same time, and the coefficient of determination (R 2 )>0.9, the model with the highest coefficient of determination or the smallest root mean square error and absolute value error is selected as the optimal parameter model. Specifically, the machine learning algorithm model written by the intelligent control unit (5) is the optimal parameter model, and hyperparameters are adjusted during its training process. The parameter adjustment methods that can be used include grid search, random search, Bayesian optimization parameter adjustment or manual parameter adjustment, which can be selected according to the size of training data and the speed of parameter adjustment.

[0008] Specifically, the halogen online detection unit (1) and the fluorescence online detection unit (2) are placed before the front ultraviolet disinfection unit (3) and the rear chlorine disinfection unit (4), respectively detecting the concentrations of iodide ions, iodate ions and bromide ions in the water, as well as the three-dimensional fluorescence level (excitation emission matrix) of dissolved organic matter in the water, and are capable of sending the data to the intelligent control unit (5).

[0009] Specifically, the front ultraviolet disinfection unit (3) can send the current front ultraviolet disinfection data to the intelligent control unit (5), and can receive the control signal transmitted by the intelligent control unit (5) to adjust the ultraviolet light intensity.

[0010] Specifically, the post-chlorination disinfection unit (4) can send current post-chlorination disinfection data to the intelligent control unit (5), and can receive a control signal transmitted by the intelligent control unit (5) to adjust the chlorine addition amount.

[0011] Specifically, the series of disinfection parameter combinations set by the intelligent control unit (5), including ultraviolet light intensity and chloramine concentration, should ensure that the factory water meets the requirements of the "National Drinking Water Quality Standard (GB 5749-2022)".

[0012] Specifically, the iodine-substituted disinfection by-product limit value set by the intelligent control unit (5) is set by the water supply plant and can be set according to the reference indicators and limits of drinking water quality in Appendix A of the "National Drinking Water Quality Standard (GB 5749-2022)", which lists iodoacetic acid not exceeding 20 μg / L, or other indicators or other values ​​can be set as needed.

[0013] Specifically, when the intelligent control unit (5) calculates that the value cannot be less than or equal to the limit set by the water supply plant, it feeds back data to the water supply plant control system (6) for early warning, and the water supply plant can take other measures or processes for further interception or processing (if any).

[0014] The present invention has the following beneficial effects:

[0015] (1) Online detection of halogen levels and dissolved organic matter three-dimensional fluorescence levels, online feedback of current disinfection parameters, and accurate characterization and fine treatment. Real-time water quality information is obtained through online detection, and fine use and regulation of it is achieved in subsequent treatment.

[0016] (2) Optimize the algorithm during the training process to capture the law. The machine learning algorithms that can be selected include but are not limited to the extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm, and random forest (RF) algorithm. The extreme gradient boosting (XGB) algorithm is preferred to effectively capture the potential nonlinear generation law.

[0017] (3) Accurately predict the production of iodine disinfection by-products to assist control and decision-making. Real-time prediction with a higher coefficient of determination and lower error is helpful for accurate quantification.

[0018] (4) Intelligently control the disinfection process to ensure effluent safety. Under intelligent control, ultraviolet disinfection parameters and chlorination disinfection parameters that produce lower iodine disinfection by-products are effectively applied to the disinfection unit, so that iodine disinfection by-products in the effluent are effectively controlled and safety risks are reduced.

[0019] (5) Effectively warn of potential risks and ensure disinfection throughout the entire process. When disinfection control cannot reduce the amount of iodine-based disinfection by-products to the expected limit, first minimize the amount of iodine-based disinfection by-products, issue a warning signal in a timely manner, and initiate other corresponding treatment measures to ensure safety.

[0020] (6) Convenient system application and integration. The smart control unit can be expanded on the existing control system of the water supply plant or directly integrated into the system, and take over the disinfection unit to achieve ease of use and convenience in the smart disinfection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the online intelligent early warning and prevention system for iodine disinfection by-products in water plants.

[0022] Figure 2 Flowchart for training the optimal parameter model for the online intelligent early warning and prevention system of iodine-containing disinfection by-products in water plants.

[0023] Figure 3 The invention is a prediction effect of iodine disinfection by-products after implementing four algorithms, namely, extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm, and random forest (RF) algorithm, with the determination coefficient as an evaluation index.

[0024] Figure 4 The invention discloses an iodine disinfection by-product prediction effect after the implementation of the preferred extreme gradient boosting (XGB) algorithm with root mean square error and absolute value error as evaluation indicators.

[0025] Figure 5 The invention compares the difference in the prediction effect of iodine-substituted disinfection by-products between the method of the invention and the method not of the invention.

[0026] Figure 6 This is a flow chart of the intelligent computing process of the online intelligent early warning and prevention and control system for iodine-substituted disinfection by-products in water plants during actual operation.

[0027] Figure 7 The figure is a comparison of iodine-substituted disinfection by-products produced during the disinfection process regulated by the method of the present invention and the method not of the present invention.

[0028] Appendix Description

[0029] Table 1 is a comparison of the actual measured values ​​of total iodinated disinfection by-products and the calculated values ​​of the embodiments of the present invention (in μg / L), and the first 30 batches are listed as examples. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the present invention, and provides a detailed implementation method and a specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0031] In the following examples, unless otherwise specified, raw materials, reagents or processing techniques are all conventional commercial products or conventional processing techniques in the art.

[0032] Embodiment The implementation content of this embodiment is the construction of the system itself in a water supply plant, and its structure is as follows Figure 1 As shown. The system structure and functions are as follows:

[0033] (1.1) The halogen online detection unit detects the concentrations of iodide ions, iodate ions and bromide ions in water online respectively, and sends the concentration data to the intelligent control unit;

[0034] (1.2) The fluorescence online detection unit performs online detection of the three-dimensional fluorescence level of dissolved organic matter in water, i.e., the excitation emission matrix;

[0035] (1.3) The front UV disinfection unit and the rear chlorination disinfection unit send the currently running disinfection parameters UV intensity and chlorination dosage to the intelligent control unit;

[0036] (1.4) The intelligent control unit predicts the iodine disinfection by-products that will be generated under the current disinfection parameters according to the written machine learning algorithm model, and compares it with the iodine disinfection by-product limit set by the water supply plant to determine whether the disinfection parameters need to be adjusted;

[0037] (1.5) Using new disinfection parameters with low risk of iodine-containing disinfection by-products to control the front UV disinfection unit and the back chlorine disinfection unit, the output of iodine-containing disinfection by-products in the effluent is reduced, thereby reducing the corresponding risk;

[0038] The system construction process of this embodiment includes building and training the machine learning model written by the intelligent control unit. The process is as follows: Figure 2 The process includes:

[0039] (2.1) A complete data set was selected, with the input variables being the concentrations of iodide, iodate, and bromide in water (in micromoles per liter), the excitation-emission matrix signal decomposed by the parallel factorization method as the fluorescence level of dissolved organic matter (in fluorescence intensity), and the pre-UV-post-chlorination disinfection parameters including UV intensity and chloramine concentration (in mJ / cm 2 , mg-Cl2 / L);

[0040] (2.2) The output variable is the total iodinated disinfection byproduct concentration (in μg / L);

[0041] (2.3) Write at least four specific algorithms, including the extreme gradient boosting (XGB) algorithm, the artificial neural network (ANN) algorithm, the support vector machine (SVM) algorithm, and the random forest (RF) algorithm;

[0042] (2.4) Splitting the complete data set into a training data set and a test data set, where the training data set is used to train the model;

[0043] (2.5) Perform hyperparameter adjustment and conduct performance testing on the test data set after the hyperparameters are initially formed;

[0044] (2.6) Finally, the coefficient of determination (R 2 )>0.9, the optimal parameter model with the highest determination coefficient or the smallest root mean square error (RMSE) and absolute error (MAE) was selected.

[0045] Example 1

[0046] This embodiment takes a water supply plant in the southeast region as the implementation background. Based on the system established in the embodiment, the extreme gradient boosting (XGB) algorithm is specifically selected for the machine learning model written by the intelligent control unit. The adjusted hyperparameters and their ranges include a decision tree depth parameter (max_depth) greater than or equal to 5 and less than or equal to 15, an estimator parameter (n_estimator) greater than or equal to 10 and less than or equal to 100, etc.

[0047] The general XGB algorithm principle (KDD'16, 785–794) is used for writing and calculation. For the measured pool, a data set with m-dimensional features and n samples is obtained.

[0048]

[0049] Set its corresponding predicted value

[0050]

[0051] in The regression tree space is:

[0052]

[0053] The loss of the tth iteration of cumulative training to the i-th instance is:

[0054]

[0055] Compute its second-order approximation:

[0056]

[0057] The multiplier term of the second-order Taylor expansion is:

[0058]

[0059]

[0060] The loss function is fitted based on the following formula:

[0061]

[0062] After simulating different disinfection treatments for more than 90 batches of raw water entering the water supply plant, the concentration of total iodinated disinfection by-products in the disinfected water (in μg / L) was determined by gas chromatography as the actual value (y i ), and the total iodinated disinfection by-product concentration (in μg / L) is calculated using the machine learning model of the intelligent control unit of the system as the predicted value Finally, we have:

[0063] Y=(y1,y2,y3,……,y 88 ,y 89 ,y 90 ) T

[0064]

[0065] The matrix Y and The specific values ​​are shown in Appendix 1.

[0066] After implementing the above process in this embodiment, the determination coefficient (R 2 ) The prediction effect is shown in Figure 3 , calculated by comparing the actual value of the test data set with the predicted value of each model:

[0067]

[0068] where y i is the actual value of iodinated disinfection by-products (μg / L), is the model prediction value (μg / L). In this implementation of the invention, the coefficient of determination (R 2 ) is 0.935≥0.9, indicating that the present invention can explain more than 90% of the variance of the predicted iodinated disinfection by-products, and the two are very close.

[0069] The prediction results obtained by examining the root mean square error (RMSE) and absolute value error (MAE) are shown in Figure 4 , calculated by comparing the actual value of the test data set with the predicted values ​​of each model:

[0070]

[0071]

[0072] where y i is the actual value of iodinated disinfection by-products (μg / L), is the model prediction value (μg / L). Under the implementation mode of the invention, the root mean square error and absolute value error of the prediction result of the total iodinated disinfection byproduct concentration of the present invention are 13.00μg / L and 9.23μg / L respectively, and the error values ​​are respectively equivalent to 4.46% and 3.17% of the maximum value of the total iodinated disinfection byproduct concentration of 291.44μg / L in the data set, or 15.36% and 10.91% of the average value of the total iodinated disinfection byproduct concentration of 84.64μg / L in the data set, indicating that the present invention has a small error degree and can accurately predict the generation of iodinated disinfection byproducts.

[0073] Further investigation of the prediction effect of Pearson correlation coefficient (r) and normalized standard deviation (ND) shows Figure 5 , calculated by comparing the actual value of the test data set with the predicted value of each model:

[0074]

[0075]

[0076] Where Y is the actual value of iodinated disinfection byproducts random variable, is a model prediction value random variable. In the implementation of the invention, the present invention achieves a prediction result of a Pearson correlation coefficient of 0.97 and a normalized standard deviation of 0.91 for the prediction result of the total iodinated disinfection byproduct concentration, and the point representing the model prediction number sequence is close to the point representing the real data sequence in the radar chart, indicating that the present invention can accurately predict the generation of iodinated disinfection byproducts in this implementation.

[0077] Example 2

[0078] This embodiment is implemented in a water supply plant in the southeast region. Based on the system established in the embodiment, an artificial neural network (ANN) algorithm is specifically selected for the machine learning model written into the intelligent control unit. The adjusted hyperparameters and their ranges include the number of hidden layers parameter (n) from 1 to 2, the number of neurons from 10 to 200, etc.

[0079] The general ANN algorithm principle (scikit-learn 1.17.7) is used for writing and calculation. The model learning function is:

[0080]

[0081] There is an activation function:

[0082]

[0083] Use the mean squared error loss function:

[0084]

[0085] In gradient descent, the gradient of the loss with respect to the weights is calculated and subtracted from it:

[0086]

[0087] where i is the iteration step and ∈ is the learning rate with a value greater than 0.

[0088] After simulating different disinfection treatments for more than 90 batches of raw water entering the water supply plant, the concentration of total iodinated disinfection by-products in the disinfected water (in μg / L) was determined by gas chromatography as the actual value (y i ), and the total iodinated disinfection by-product concentration (in μg / L) is calculated using the machine learning model of the intelligent control unit of the system as the predicted value Finally, we have:

[0089] Y=(y1,y2,y3,……,y 88 ,y 89 ,y 90 ) T

[0090]

[0091] The matrix Y and The specific values ​​are shown in Appendix 1.

[0092] After implementing the above process in this embodiment, the determination coefficient (R 2 ) The prediction effect is shown in Figure 3 , calculated by comparing the actual value of the test data set with the predicted values ​​of each model:

[0093]

[0094] where y i is the actual value of iodinated disinfection by-products (μg / L), is the model prediction value (μg / L). In this implementation of the invention, the coefficient of determination (R 2 ) is 0.910≥0.9, indicating that the present invention can explain more than 90% of the variance of the predicted iodinated disinfection by-products, and the two are very close.

[0095] Further investigation of the prediction effect of Pearson correlation coefficient (r) and normalized standard deviation (ND) shows Figure 5 , calculated by comparing the actual value of the test data set with the predicted values ​​of each model:

[0096]

[0097]

[0098] Where Y is the actual value of iodinated disinfection byproducts random variable, is a model prediction value random variable. In the implementation of the invention, the present invention achieves a prediction result of a Pearson correlation coefficient of 0.96 and a normalized standard deviation of 0.88 for the prediction result of the total iodinated disinfection byproduct concentration, and the point representing the model prediction number sequence is close to the point representing the real data sequence in the radar chart, indicating that the present invention can accurately predict the generation of iodinated disinfection byproducts in this implementation.

[0099] Example 3

[0100] This embodiment takes a water supply plant in the southeast region as the implementation background. Based on the system established by the embodiment, a support vector machine (SVM) algorithm is specifically selected for the machine learning model written by the intelligent control unit. The adjusted hyperparameters and their ranges include a polynomial kernel order parameter (degree) greater than or equal to 3 and less than or equal to 5, a regularization parameter (C) greater than or equal to 0.9 and less than or equal to 1.0, etc.

[0101] Using the general SVM algorithm principle (Machine Learning.20,273–297,1995) for writing and calculation, we can get the optimal hyperplane in the feature space:

[0102] w0·z+b0=0

[0103] The optimal hyperplane weight w0 in the feature space is a linear combination of feature vectors:

[0104]

[0105] The linear determination function I(z) in feature space is as follows:

[0106]

[0107] Maximize margin:

[0108]

[0109] Under the constraints:

[0110] y i (X i w+b)≥1-ξ i ,i=1,…,l

[0111] ξ i ≥0,i=1,…,l

[0112] The Lagrangian function at this time is:

[0113]

[0114] Maximize the function to obtain the hyperplane solution:

[0115]

[0116] After simulating different disinfection treatments for more than 90 batches of raw water entering the water supply plant, the concentration of total iodinated disinfection by-products in the disinfected water (in μg / L) was determined by gas chromatography as the actual value (y i ), and the total iodinated disinfection by-product concentration (in μg / L) is calculated using the machine learning model of the intelligent control unit of the system as the predicted value Finally, we have:

[0117] Y=(y1,y2,y3,……,y 88 ,y 89 ,y 90 ) T

[0118]

[0119] The matrix Y and The specific values ​​are shown in Appendix 1.

[0120] After implementing the above process in this embodiment, the determination coefficient (R 2 ) The prediction effect is shown in Figure 3 , calculated by comparing the actual value of the test data set with the predicted values ​​of each model:

[0121]

[0122] where y i is the actual value of iodinated disinfection by-products (μg / L), is the model prediction value (μg / L). In this implementation of the invention, the coefficient of determination (R 2 ) is 0.908≥0.9, indicating that the present invention can explain more than 90% of the variance of the predicted iodinated disinfection by-products, and the two are very close.

[0123] Further investigation of the prediction effect of Pearson correlation coefficient (r) and normalized standard deviation (ND) shows Figure 5 , calculated by comparing the actual value of the test data set with the predicted values ​​of each model:

[0124]

[0125]

[0126] Where Y is the actual value of iodinated disinfection byproducts random variable, is a model prediction value random variable. In the implementation of the invention, the present invention achieves a prediction result of a Pearson correlation coefficient of 0.95 and a normalized standard deviation of 0.92 for the prediction result of the total iodinated disinfection byproduct concentration, and the point representing the model prediction number sequence is close to the point representing the real data sequence in the radar chart, indicating that the present invention can accurately predict the generation of iodinated disinfection byproducts in this implementation.

[0127] Example 4

[0128] This embodiment takes a water supply plant in the southeast region as the implementation background. Based on the system established in the embodiment, the random forest (RF) algorithm is specifically selected for the machine learning model written by the intelligent control unit. The adjusted hyperparameters and their ranges include the decision tree depth parameter (max_depth) greater than or equal to 5 and less than or equal to 20, the estimator parameter (n_estimator) greater than or equal to 10 and less than or equal to 100, etc.

[0129] The general RF algorithm (Machine Learning.45, 5–32, 2001) is used for writing and calculation, where for the sample data set input feature X, there is an RF estimator:

[0130] h=h(X,Θ)

[0131] Where Θ is a random vector. For sample input features X and output Y, there is a mean squared generalization error E:

[0132] E=E X,Y (Yh(X)) 2

[0133] By k )} Take the average to get the predicted value, we have:

[0134] E X,Y (Y-av k h(X,Θ k )) 2 →E X,Y (YE Θ h(X,Θ k )) 2 ,k→∞

[0135] The average generalization error of the tree is:

[0136] PE * (tree)=E Θ E X,Y (Yh(X,Θ)) 2

[0137] Then the RF average generalization error is:

[0138] PE * (forest)=E X,Y [E Θ (Yh(X,Θ))] 2

[0139] And there are:

[0140] PE * (forest)≤PE * (tree)

[0141] After simulating different disinfection treatments for more than 90 batches of raw water entering the water supply plant, the concentration of total iodinated disinfection by-products in the disinfected water (in μg / L) was determined by gas chromatography as the actual value (y i ), and the total iodinated disinfection by-product concentration (in μg / L) is calculated using the machine learning model of the intelligent control unit of the system as the predicted value Finally, we have:

[0142] Y=(y1,y2,y3,……,y 88 ,y 89 ,y 90 ) T

[0143]

[0144] The matrix Y and The specific values ​​are shown in Appendix 1.

[0145] After implementing the above process in this embodiment, the determination coefficient (R 2 ) The prediction results are shown in Figure 3 , calculated by comparing the actual value of the test data set with the predicted value of each model:

[0146]

[0147] where y i is the actual value of iodinated disinfection by-products (μg / L), is the model prediction value (μg / L). In this implementation of the invention, the coefficient of determination (R 2 ) is 0.907≥0.9, indicating that the present invention can explain more than 90% of the variance of the predicted iodinated disinfection by-products, and the two are very close.

[0148] Further investigation of the prediction effect of Pearson correlation coefficient (r) and normalized standard deviation (ND) shows Figure 5 , calculated by comparing the actual value of the test data set with the predicted value of each model:

[0149]

[0150]

[0151] Where Y is the actual value of iodinated disinfection byproducts random variable, is a model prediction value random variable. In the implementation of the invention, the present invention achieves a prediction result of a Pearson correlation coefficient of 0.96 and a normalized standard deviation of 0.82 for the prediction result of the total iodinated disinfection byproduct concentration, and the point representing the model prediction number sequence is close to the point representing the real data sequence in the radar chart, indicating that the present invention can accurately predict the generation of iodinated disinfection byproducts in this implementation.

[0152] Example 5

[0153] This embodiment is based on a water supply plant in the southeast region. Based on the system established in the embodiment, the extreme gradient boosting (XGB) algorithm is specifically selected for the machine learning model written in the intelligent control unit. Under the simulated ultraviolet dose of 1500mJ / cm 2 Under the current disinfection parameters of simulating the addition of chloramine at a rate of 5 mg-Cl2 / L, the monitoring limit of total iodinated disinfection byproducts in the water supply plant is set at 100 μg / L, and Figure 6 The process of the present invention is carried out to regulate the disinfection, and the yield of the iodine disinfection by-product is obtained as follows Figure 7 In this implementation, the present invention changes the disinfection parameter to a UV dose of 0 mJ / cm by intelligent regulation. 2 With the addition of chloramine at a rate of 5 mg-Cl2 / L, the final total concentration of dichloroiodomethane, monochlorodiiodomethane, triiodomethane, monoiodoacetic acid and diiodoacetic acid was only 61.38 μg / L, which was far below the limit of 100 μg / L, indicating that the present invention can achieve effective control of iodinated disinfection by-products.

[0154] Comparative Example 1

[0155] Select the multiple linear regression algorithm as a comparative example, keep other application conditions the same as the above example, according to Figure 2 The same training is performed by the process shown in the figure, and the difference between the prediction effect of the comparative example and the prediction effect of the embodiment of the present invention is shown in FIG. Figure 5. The present invention under four different implementation modes such as extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm, random forest (RF) algorithm, etc., respectively achieves the prediction effect of Pearson correlation coefficient reaching 0.97, 0.96, 0.95, 0.96, and normalized standard deviation reaching 0.91, 0.88, 0.92, 0.82, while the comparative example not using the present invention can only achieve the prediction result of Pearson correlation coefficient 0.71 and normalized standard deviation 0.68, and its distance from the real data in the radar chart far exceeds the distance between the present invention and the real data. This shows that under this condition, the present invention is better than the prediction effect of iodine disinfection by-products realized by multiple linear regression algorithm.

[0156] Comparative Example 2

[0157] The conventional constant disinfection parameter control method was selected as a comparative example, and other application conditions were kept the same as those in the above embodiment. The raw water with the same water quality conditions as in Example 5 was disinfected. The simulated UV dose was 1500mJ / cm 2 Under the current disinfection parameter of adding chloramine at 5 mg-Cl2 / L, the monitoring limit of total iodinated disinfection by-products was set at 100 μg / L, and the comparison of iodinated disinfection by-products was obtained. Figure 7 As shown. The final total concentration reached 206.25μg / L, more than twice the limit of 100μg / L, which is 336.02% of the present invention. The present invention achieved a reduction of up to 70.2% of iodine disinfection by-products through intelligent regulation. This shows that under this condition, the ability of the present invention to reduce risks is better than the constant disinfection parameter control method.

[0158] Table 1

[0159]

[0160]

Claims

1. An online intelligent early warning and prevention system for iodine disinfection by-products in a water supply plant, characterized in that: The system comprises a halogen online detection unit (1), a fluorescence online detection unit (2), a pre-ultraviolet disinfection unit (3), a post-chlorination disinfection unit (4), and an intelligent control unit (5), and each unit is integrated and connected with a water plant control system (6); when the system treats iodine-containing water, first, the halogen online detection unit (1) respectively detects the concentrations of iodine ions, iodate ions and bromide ions in the water online, and sends the concentration data to the intelligent control unit (5); at the same time, the fluorescence online detection unit (2) detects the three-dimensional fluorescence level of dissolved organic matter in the water, that is, the excitation emission matrix, online, and sends the result to the intelligent control unit (5); at the same time, the pre-ultraviolet disinfection unit (3) and the post-chlorination disinfection unit (4) that have not yet been disinfected send the currently running disinfection parameters of ultraviolet light intensity and chlorination dosage to the intelligent control unit (5), and the intelligent control unit (5) performs online detection according to the machine learning that has been written. The learning algorithm model predicts the iodine disinfection by-products that will be generated under the current disinfection parameters, and compares them with the iodine disinfection by-product limit values ​​that have been set by the water supply plant to determine whether the disinfection parameters need to be adjusted, and then controls the front ultraviolet disinfection unit (3) and the post-chlorination disinfection unit (4) with the new disinfection parameters with low iodine disinfection by-product generation risk, so that the output of iodine disinfection by-products in the effluent is reduced, thereby reducing the corresponding risk; the machine learning algorithm model written into the intelligent control unit (5) is an extreme gradient boosting (XGB) algorithm, an artificial neural network (ANN) algorithm, a support vector machine (SVM) algorithm or a random forest (RF) algorithm. At the same time, the model used should be the optimal parameter model trained by the water supply plant. The optimal parameter model is generated by adjusting the hyperparameters on the training data set after segmenting the complete data set, and performing a test set model test to achieve a determination coefficient R on the test data set. 2 The model with a value of >0.9 is finally applied. The input variables of the complete training data set include the concentrations of iodine ions, iodate ions and bromide ions in water, the fluorescence level of dissolved organic matter, and the pre-ultraviolet-post-chlorination disinfection parameters. The output variable is the level of iodine-containing disinfection by-products under the corresponding input conditions. When applied, the intelligent control unit (5) receives the data from the halogen online detection unit (1), the fluorescence online detection unit (2), the pre-ultraviolet disinfection unit (3) and the post-chlorination disinfection unit (4), and performs predictive calculation and logical judgment. The process is to first predict the level of iodine-containing disinfection by-products under the current disinfection parameters. If it is less than the iodine-containing disinfection by-product limit set by the water plant, the current disinfection parameters are used to calculate the level of iodine-containing disinfection by-products. The disinfection parameter continues to run. If it is greater than the iodine disinfection by-product limit value set by the water plant, the series of disinfection parameters are cyclically input into the model for prediction, and the disinfection parameters of the front ultraviolet disinfection unit (3) and the rear chlorination disinfection unit (4) are reset to perform disinfection with the disinfection parameter control whose prediction result is less than the iodine disinfection by-product limit value set by the water plant. If all disinfection parameters have no prediction results less than the iodine disinfection by-product limit value set by the water plant after the cyclic input, the disinfection parameters of the front ultraviolet disinfection unit (3) and the rear chlorination disinfection unit (4) are reset to perform disinfection with the disinfection parameter control whose iodine disinfection by-product generation level is the minimum value, and the data is fed back to the water plant control system (6) for early warning.

2. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The machine learning algorithm model written into the intelligent control unit (5) is an optimal parameter model. The generation and selection of the optimal parameter model includes the following steps: ① Establishing a model with historical monitoring data of the water supply plant using the system, or open source water quality experimental data. The input variables of the complete data set used in the training process include three types of water halogen levels, dissolved organic matter fluorescence levels, and pre-ultraviolet-post-chlorination disinfection parameters. The halogen levels are specifically iodine ion, iodate ion and bromide ion concentrations. The dissolved organic matter fluorescence level is a decomposed excitation emission matrix signal. The decomposition method is selected from the parallel factor method, the principal component analysis method or the pre-machine learning method. The pre-ultraviolet-post-chlorination disinfection parameters include ultraviolet light intensity and chloramine concentration; ② Input The output variable is the level of iodinated disinfection by-products under the corresponding input conditions, wherein the level of iodinated disinfection by-products uses the total iodinated disinfection by-product concentration according to the monitoring needs, or uses the overall index of total iodinated trihalomethane and total iodinated haloacetic acid concentration according to the monitoring needs, or uses the concentration of a single or several specific iodinated disinfection by-products. The total iodinated disinfection by-products include triiodomethane, monochlorodiiodomethane, dichloromonoiodomethane, monobromodiiodomethane, dibromomonoiodomethane, chlorobromoiodomethane, as well as monoiodoacetic acid, diiodoacetic acid or triiodoacetic acid; ③ Use the extreme gradient boosting (XGB) algorithm, artificial neural network (ANN) algorithm, support vector machine (SVM) algorithm or random forest (RF) algorithm for training, adjustment and testing at the same time, and the determination coefficient R 2 Among the models with a coefficient of determination greater than 0.9, the model with the highest coefficient of determination or the smallest root mean square error or absolute value error is selected as the optimal parameter model.

3. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The halogen online detection unit (1) and the fluorescence online detection unit (2) are placed before the front ultraviolet disinfection unit (3) and the rear chlorination disinfection unit (4), respectively detecting the concentrations of iodide ions, iodate ions and bromide ions in the water, and determining the three-dimensional fluorescence level of dissolved organic matter in the water by detecting the excitation emission matrix, and are capable of sending the data to the intelligent control unit (5).

4. According to the online intelligent early warning and prevention system for iodine disinfection by-products in water supply plants as claimed in claim 1, the machine learning algorithm model written into the intelligent control unit (5) is an extreme gradient boosting (XGB) algorithm.

5. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The front ultraviolet disinfection unit (3) can send current front ultraviolet disinfection data to the intelligent control unit (5), and can receive a control signal transmitted by the intelligent control unit (5) to adjust the ultraviolet light intensity.

6. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The post-chlorination disinfection unit (4) can send current post-chlorination disinfection data to the intelligent control unit (5), and can receive a control signal transmitted by the intelligent control unit (5) to adjust the chlorine addition amount.

7. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The machine learning algorithm model written into the intelligent control unit (5) is an optimal parameter model, and hyperparameters are adjusted during its training process. The parameter adjustment methods that can be used include grid search, random search, Bayesian optimization parameter adjustment or manual parameter adjustment, which can be selected based on the size of training data and the parameter adjustment speed.

8. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The disinfection parameters set by the intelligent control unit (5) include ultraviolet light intensity and chloramine concentration.

9. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: The iodine disinfection by-product limit value set by the intelligent control unit (5) is set by the water supply plant.

10. The online intelligent early warning and prevention system for iodine disinfection byproducts in water supply plants according to claim 1, characterized in that: When the intelligent control unit (5) calculates that the value cannot be less than the limit set by the water supply plant, it feeds back data to the water supply plant control system (6) for early warning.

11. An online intelligent early warning and prevention method for iodine disinfection by-products in a water supply plant, characterized in that: The method is implemented by the online intelligent early warning and prevention and control system for iodine-substituted disinfection by-products in water supply plants according to any one of claims 1 to 10.

Images